Generate Videos
November 17, 2025 (November 20, 2026)
Table of contents
- Model Capabilities
- Content Moderation
- Why a generation failed
- Request Headers
- Request Body
- Responses
- Model
- Missing media URLs
- Examples
- Try It
Generate videos using Google Flow AI models (Omni 1.2 Flash, Omni 1.1 Flash, Veo 3.1 Quality, Veo 3.1 Fast, Veo 3.1 Lite, Veo 3.1 Lite Lower Priority) from text prompts with optional start/end frames, reference images, reference video (Omni V2V edit), and voice narration. Videos are returned as signed URLs ready for download.
Omni 1.2 Flash (omni-flash-1.2) makes clips of any whole number of seconds from 3 to 15, at 360p or 720p, and is the model POST /videos/extend extends with. Google released it on November 5, 2026.
Google is giving Omni 1.2 Flash to accounts over several days from November 5, 2026. Until an account has it, a request for omni-flash-1.2 on that account returns 400 straight away, before any captcha is solved. When the load balancer picks the account (no email), accounts that don’t have the model yet are skipped.
This endpoint features dynamic concurrency management. Video generation typically completes within 60-180 seconds depending on the model and mode. The endpoint automatically polls for completion and returns the final result.
Veo 3.1 Lite Lower Priority is available only to Google AI Ultra $199 subscribers — no credit cost but lower generation priority. The cheaper Ultra $99 plan does not include it. Since 2026-09-23 Google no longer offers it to invited members of an Ultra $199 family plan — only the family manager’s account keeps it. A member account still shows its Ultra credits, but the model is missing from its model list at flow.google.com. The API checks that model list, so a request for this model on a member account returns 400 straight away, before any captcha is solved. When the load balancer picks the account (no email, and no input image, reference, voice or character that ties the request to one account), member accounts are skipped for this model automatically. If Google still refuses the model on an account, the request returns 403 with PUBLIC_ERROR_MODEL_ACCESS_DENIED and the account is skipped for this model for an hour — see Load Balancing Algorithm.
Model Capabilities
Two model families: Veo 3.1 (four variants) and Omni Flash (two versions, Omni 1.2 Flash and Omni 1.1 Flash).
The omni-flash model value is Omni 1.1 Flash, the label the Flow interface has used since Google’s August 26, 2026 update. omni-flash-1.2 is Omni 1.2 Flash, a separate model with its own lengths and prices. The current label of each is shown as displayName on GET /accounts/email.
Models
Veo figures: Google Flow — Credits and pricing. Omni figures measured against the live API.
Omni 1.2 Flash is priced per second, the same on every plan: the first 10 seconds at the base rate and every second past 10 at twice that. An edit costs the same as a generation of the edited length.
| Model | Generation type(s) | Credits per generation |
|---|---|---|
Veo 3.1 - Liteveo-3.1-lite | 4s, 6s, 8s videos | Non-Ultra: 10 Ultra: 5 |
Veo 3.1 - Lite Lower Priorityveo-3.1-lite-low-priority | 4s, 6s, 8s videos | Ultra $199 only: 0 (lower priority) |
Veo 3.1 - Fast (default until 2026-12-05)veo-3.1-fastretiring 2026-12-05, then runs as veo-3.1-lite | 4s, 6s, 8s videos | Non-Ultra: 20 Ultra: 10 |
Veo 3.1 - Qualityveo-3.1-qualityretiring 2026-12-05, then runs as omni-flash-1.2 at 720p | 4s, 6s, 8s videos | 100 |
Omni 1.2 Flashomni-flash-1.2 | 3s to 15s videos at 720p Edit, Extend | Every plan (Free: up to 10s, no extend): 3 per second up to 10s, 6 per second above 3s: 9 · 8s: 24 · 10s: 30 · 15s: 60 |
Omni 1.2 Flashomni-flash-1.2 + resolution: 360p | 3s to 15s videos at 360p Edit, Extend | Every plan (Free: up to 10s, no extend): 1 per second up to 10s, 2 per second above 3s: 3 · 8s: 8 · 10s: 10 · 15s: 20 |
Omni 1.1 Flashomni-flashretiring 2026-12-05, then runs as omni-flash-1.2 | 4s, 6s, 8s, 10s videos at 720p (does not include edits to uploaded & generated videos) | Plus / Pro / Ultra: 4s: 7 6s: 10 8s: 12 10s: 15 |
Omni 1.1 Flashomni-flash + resolution: 360pretiring 2026-12-05 | 4s, 6s, 8s, 10s videos at 360p | Plus / Pro / Ultra: 4s: 4 6s: 5 8s: 6 10s: 7 |
Omni 1.1 Flashomni-flash + referenceVideo_1retiring 2026-12-05 | Edit uploaded & generated videos of all lengths | Plus / Pro / Ultra: 20 at 720p, 10 at 360p |
Requests for a retiring model keep working after December 5, 2026: from that day a request without model runs as veo-3.1-lite, the new default, and omni-flash at 720p costs about twice the credits (at 360p about the same). The job then shows the model that ran in model.
Aspect ratio (aspectRatio)
| Model | Supported values |
|---|---|
| All Veo variants | landscape (default), portrait |
omni-flash-1.2, omni-flash | landscape (default), portrait |
Duration (duration, sec)
| Mode / Model | Supported |
|---|---|
| Veo T2V / I2V / I2V-FL | 4, 6, 8 (default) — 4/6 are Ultra-only |
| Veo R2V | 8 only |
omni-flash-1.2 T2V / R2V / I2V | any whole number from 3 to 15, 8 (default) — 11 to 15 on paid plans only |
omni-flash-1.2 I2V-FL | any whole number from 3 to 10, 8 (default) |
omni-flash-1.2 V2V edit | not accepted — output matches input trim window (max 15 s, set via endFrameIndex_1, max 10 s on a free plan) |
omni-flash T2V / R2V / I2V / I2V-FL | 4, 6, 8 (default), 10 |
omni-flash V2V edit | not accepted — output matches input trim window (max 10 s, set via endFrameIndex_1) |
References
| Reference | Parameter(s) | Veo (all variants) | omni-flash-1.2, omni-flash |
|---|---|---|---|
| Start frame (I2V) | startImage | referenceImage_*, no character_*, no referenceAudio_*) | |
| Start + end frames (I2V-FL) | startImage + endImage | referenceImage_*, no character_*, no referenceAudio_*) | |
| Image refs (R2V / Ingredients) | referenceImage_1..10 | up to 3 (slots _1..3); not on veo-3.1-quality | up to 10 on omni-flash-1.2, 7 on omni-flash (R2V) / up to 5 (V2V edit) |
| Characters (R2V / Ingredients) | character_1..10 (from POST /characters) | 3 total with referenceImage_* — Veo fast/lite/lite-low-priority only, 8 s only, not on veo-3.1-quality | 10 total with referenceImage_* on omni-flash-1.2, 7 on omni-flash (R2V) |
| Voice narration | referenceAudio_1..5 (POST /voices refs OR system voice names) | 1 (slot _1, requires an image ref or character, R2V only); not on veo-3.1-quality | up to 5 (R2V, requires an image ref or character) / up to 3 (V2V edit) |
| Source video (V2V edit) | referenceVideo_1 (+ optional startFrameIndex_1 / endFrameIndex_1) | — | omni-flash-1.2, 240 frames = 10 s on omni-flash) |
Content Moderation
If your generation is moderated, retrying with the same prompt often succeeds on subsequent attempts — moderation decisions can vary between requests. Modifying the prompt, removing reference images, or swapping models can also help, as each has different content-moderation behavior.
One case does not improve on retry. IP_PROHIBITED means Google’s intellectual-property classifier flagged an input image as containing copyrighted, branded or recognizable content, and since the image is unchanged a retry gets flagged again. Replace the reference image instead — original photos of non-famous subjects pass where celebrity photos, film stills, product shots and copyrighted characters do not. Despite how it reads, the “IP” here is intellectual property and has nothing to do with network addresses.
Why a generation failed
When every operation in a job fails, the top-level error reads All operations failed for all of them. The reason is in response.failureReasons, a de-duplicated list of everything Google said about the failed media.
Two kinds of string arrive in it, and a job can carry either or both. Terminal codes name what went wrong, and the ones seen most are PUBLIC_ERROR_UNSAFE_GENERATION, PUBLIC_ERROR_PROMINENT_PEOPLE_FILTER_FAILED, PUBLIC_ERROR_AUDIO_FILTERED, PUBLIC_ERROR_MINOR, PUBLIC_ERROR_SEXUAL, PUBLIC_ERROR_DANGER_FILTER, PUBLIC_ERROR_IP_INPUT_IMAGE and, on V2V edits, PUBLIC_ERROR_SPEECH_EDIT. Classifier labels name which filter fired, such as IP_PROHIBITED. The list is Google’s, not ours, so treat it as open-ended and match on substrings rather than on an exhaustive set.
PUBLIC_ERROR_SPEECH_EDIT comes only from V2V edits (referenceVideo_1), and it is the most common named reason there: about a quarter of failed edits carry it. Google refuses the edit because of the speech in the clip and gives no detail beyond the code. The name points at an edit that changes what someone in the video says, so try a prompt that leaves the dialogue as it is, or a source clip without speech.
The field is absent when Google returned no reason at all, which is common enough to code for — around half of V2V edit failures arrive with nothing attached. An absent failureReasons is not an error in itself, and those jobs are usually worth one retry.
A job that failed before any operation started — a rejected request, a moderated prompt — never reaches this path. Those return the Google error verbatim instead, as shown in the 400 responses above.
Videos that failed on their sound
PUBLIC_ERROR_AUDIO_FILTERED, with the label AUDIO_GENERATION_FILTERED, means the picture was fine but Google’s audio filter rejected the sound generated with it. Every model makes sound along with the video, even for a prompt with no dialogue (ambient noise, music, effects), and Google checks that sound separately from the picture. When the filter rejects it, Google fails the whole video by default and charges no credits for it.
It is the most common failure on veo-3.1-lite and the second most common on veo-3.1-lite-low-priority: about 1 in 5 veo-3.1-lite videos and 1 in 12 veo-3.1-lite-low-priority videos fail this way, about 1 in 25 on veo-3.1-fast, and fewer than 1 in 100 on the Omni models.
What you can do:
- If the video can do without its sound (b-roll, backgrounds, clips you put music or a voice-over on later), send
returnSilentVideos: true. Such a video then comes back without sound instead of failing, and Google charges its credits like any other video. This is the “Return silent videos” setting in Google Flow’s own settings, applied per request. - If you need the sound, send the request again: each attempt generates new sound, so the filter judges it anew.
- An Omni model (
omni-flash-1.2,omni-flash) hardly ever fails this way.
https://api.useapi.net/v1/google-flow/videos
Request Headers
Authorization: Bearer {API token}
Content-Type: application/json
# Alternatively you can use multipart/form-data
# Content-Type: multipart/form-data
API tokenis required, see Setup useapi.net for details.
Request Body
{
"prompt": "A serene mountain landscape at sunset with camera slowly panning right",
"model": "veo-3.1-fast",
"aspectRatio": "landscape",
"duration": 8,
"count": 2
}
Parameters are grouped below. For per-model limits and modes see Model Capabilities above.
Core
promptis required, text description for video generation.modelis optional, the AI model to use for video generation (default:veo-3.1-fast, andveo-3.1-litefrom December 5, 2026, when Google retires Veo 3.1 Fast). Supported values:omni-flash-1.2,omni-flash,veo-3.1-quality,veo-3.1-fast,veo-3.1-lite,veo-3.1-lite-low-priority. See Model Capabilities.aspectRatiois optional, output video aspect ratio (default:landscape). Supported values:landscape,portrait, for every model.durationis optional, output video length in seconds (default:8).omni-flash-1.2takes any whole number from3to15(3to10withstartImage+endImage, and up to10on a free Google plan).omni-flashtakes4,6,8,10. Veo takes4,6,8. See Model Capabilities. Not accepted by a V2V edit (output length matches the input trim window).resolutionis optional, output resolution (default:720p). Supported values:360p,720p. Omni models only — Veo publishes no 360p variant, so360pon a Veo model is rejected rather than silently generating 720p at full price. Onomni-flash-1.2a 360p generation costs a third of the same request at 720p, onomni-flashroughly half. A 360p clip can be promoted afterwards with POST /videos/upscale.countis optional, number of video variations to generate (1-4, default:1).
Account selection
-
emailis optional, the Google Flow account to use.When only one account is configured, the API will automatically use that account.
With multiple accounts configured, omitting the email parameter triggers automatic load balancing based on video generation job statistics to select the healthiest account.
If
startImage,endImage,referenceImage_*, orreferenceVideo_1are provided, theemailparameter can be omitted — the API will automatically use the same account where the references were uploaded.
Image references
startImageis optional, mediaGenerationId from POST /assets/emailfor I2V mode (video starts with this frame). Supported on Veo and both Omni models, all of which also acceptendImage.endImageis optional, mediaGenerationId from POST /assets/emailfor I2V-FL mode (video ends with this frame, requiresstartImage). End-frame-only is not supported. On the Omni models the two frames are the only accepted input, soreferenceImage_*,character_*andreferenceAudio_*are rejected alongside them.referenceImage_1toreferenceImage_10are optional, mediaGenerationId from POST /assets/emailfor R2V mode. Slots_1.._3work on Veo (max 3, notveo-3.1-quality) and both Omni models, slots_4.._7on the Omni models only, slots_8.._10onomni-flash-1.2only. Cannot be combined withstartImage/endImage.
Characters
character_1tocharacter_10are optional, character reference-ids from POST /characters. Triggers R2V (Ingredients) mode just likereferenceImage_*does. Each character’s image count rolls into the model’s image-ref budget (see Model Capabilities › References):- Veo 3.1 Fast / Lite / Lite Lower Priority (
veo-3.1-fast,veo-3.1-lite,veo-3.1-lite-low-priority) — characters supported in 8 s only;referenceImages.length + Σ(character.imgs) ≤ 3. - Omni 1.2 Flash and Omni 1.1 Flash (
omni-flash-1.2,omni-flash) — characters supported in all durations.referenceImages.length + Σ(character.imgs) ≤ 10onomni-flash-1.2,≤ 7onomni-flash. - Veo 3.1 Quality (
veo-3.1-quality) — characters NOT supported (no Ingredients mode).
- Veo 3.1 Fast / Lite / Lite Lower Priority (
Voice narration
referenceAudio_1toreferenceAudio_5are optional. Each slot accepts either a system voice name (case-insensitive, one of the 30 presets) or a user voice from POST /voices. Per-model limits are in the References table above.
System voice presets (30 names):
[
"Achird", "Achernar", "Algieba", "Algenib", "Alnilam",
"Aoede", "Autonoe", "Callirrhoe", "Charon", "Despina",
"Enceladus", "Erinome", "Fenrir", "Gacrux", "Iapetus",
"Kore", "Laomedeia", "Leda", "Orus", "Puck",
"Pulcherrima", "Rasalgethi", "Sadachbia", "Sadaltager", "Schedar",
"Sulafat", "Umbriel", "Vindemiatrix", "Zephyr", "Zubenelgenubi"
]
System voice samples: https://www.gstatic.com/aitestkitchen/voices/samples/{Name}.wav
To list user (custom) voices on an account: GET /voices?email=…&source=user. To preview a single voice (system or user): GET /voices/ref.
Silent videos
returnSilentVideosis optional (default:false). Every model generates sound with the video, and Google checks that sound with its own audio filter. When the filter rejects it,falsefails the video withPUBLIC_ERROR_AUDIO_FILTEREDand Google charges nothing for it.truereturns the video without sound, charged like any other video. It is the same as “Return silent videos” in Google Flow’s settings, applied to this request only, whatever the account’s own setting says. See Videos that failed on their sound.
Video reference (Omni V2V edit)
referenceVideo_1is optional (Omni models only), mediaGenerationId of a video uploaded via POST /assets/email, or of a video this API generated or edited earlier. Switches the request into V2V edit mode. Cannot be combined withstartImage/endImage.startFrameIndex_1is optional (Omni V2V only, paired withreferenceVideo_1), trim start on a 24 fps virtual timeline. Range: 0-359 onomni-flash-1.2, 0-239 onomni-flash. Default:0.endFrameIndex_1is optional (Omni V2V only, paired withreferenceVideo_1), trim end on a 24 fps virtual timeline. Range: 1-360 onomni-flash-1.2(360 = 15 s), 1-240 onomni-flash(240 = 10 s). Default: the end of the clip, or the model’s maximum for longer clips. Must be greater thanstartFrameIndex_1. On a free Google plan anomni-flash-1.2edit covers at most 10 s (240 frames).
An edit works on a slice of the input video, and startFrameIndex_1 / endFrameIndex_1 pick that slice. Google counts 24 frames per second, so frame 24 is one second in, frame 120 is five seconds in, and frame 240 is ten seconds in. The output is that slice re-rendered with your edit, so it is exactly as long as the slice. There is no duration parameter on an edit. On omni-flash-1.2 an edit costs the same as a generation of the slice’s length (a 3 s slice at 360p costs 3 credits).
The table shows omni-flash. On omni-flash-1.2 the same rules hold with a 15 s (360 frame) maximum, so a 13 s clip is edited whole.
| Input clip | You send | Slice used | Output |
|---|---|---|---|
| 10 s | nothing | frames 0 to 240 | 10 s |
| 10 s | endFrameIndex_1: 120 | frames 0 to 120 | 5 s |
| 13 s | nothing | frames 0 to 240 | 10 s, the last 3 s of the input are not used |
| 8 s | nothing | frames 0 to 192 | 8 s |
An edit does not add time to a video. The motion follows the source closely: in our tests the edit reproduced the source’s first and last frames almost exactly, and a prompt asking it to begin elsewhere was ignored. Feeding the output back in as the next referenceVideo_1 with a new line to say gives you a new take of the same seconds, with the voice and the gestures carried over from the source.
To make a clip longer, use POST /videos/extend, which adds up to 10 seconds at a time with Omni 1.2 Flash and returns one continuous clip. Chaining edits is the other way to get a longer piece out of an Omni clip, and the joins can be made invisible with one precaution at the start. Because every edit opens and closes on the same frames as its source, the only thing that shows at a join is the difference between the source’s own first and last frame. Remove it at the seed: generate a still of the presenter at rest with POST /images, make the first clip with that still as both startImage and endImage, and every edit chained from it opens and closes on that same frame. Join the run with POST /videos/concatenate and the cuts land still-to-still. The quiet beat that pins each clip to the still also shows at a join as a hold, about half a second before the presenter speaks and a third of a second of settling after, so trim it there: roughly trimEnd: 0.375 on every clip but the last and trimStart: 0.458 on every clip but the first leaves two frames of rest across each cut. Measured on a 4-second talking head, the joins were then indistinguishable from the motion inside the clips and the longest hold was an eighth of a second.
Google’s V2V moderation is non-deterministic. The exact same body can land on a content-safety reject (IDENTIFIABLE_PERSON_SAFETY, CHILD_SAFETY, etc.) on one attempt and a clean success on the next. A simple resubmit usually clears it — no input changes needed.
Inline @-mention markers (optional)
You can place @-markers inside prompt that resolve to the body parameters you supply. Markers are case-insensitive and opt-in — a prompt with no markers behaves exactly like before.
Supported markers (one per body slot):
| Marker | Resolves to |
|---|---|
@character_1..10 | matching character_N body param |
@referenceImage_1..10 | matching referenceImage_N body param |
@referenceAudio_1..5 | matching referenceAudio_N body param |
@referenceVideo_* | NOT supported inline — pass referenceVideo_1 as a body field. The API returns 400 if it appears in prompt. |
Rules:
- Each marker MUST have a matching body slot supplied. Otherwise the API returns 400 with
'character_N' was not provided in the request body. - A slot referenced inline still goes into the upstream reference array — markers are positional/grounding info, not a replacement for the body params.
- Same slot may be inlined multiple times in one prompt; the upstream reference array de-duplicates by id.
- Body params supplied but NOT inlined still flow to the upstream reference array (unchanged behavior).
- Unknown strings like
@gmail.comor@anyword_1(no_<digit>after a recognized slot name) pass through as literal prompt text.
Example:
{
"model": "omni-flash",
"prompt": "@character_1 sitting on @referenceImage_1, @character_2 standing beside, @referenceImage_2 visible in the background, smooth camera dolly",
"character_1": "user:123-email:...-character:...-imgs:1",
"character_2": "user:123-email:...-character:...-imgs:1",
"referenceImage_1": "user:123-email:...-image:...",
"referenceImage_2": "user:123-email:...-image:..."
}
Async & callbacks
asyncis optional, enables fire-and-forget mode (default:false). Whentrue, returns immediately with201 Createdand job metadata. Poll GET /jobs/jobIdfor completion status. Useful for avoiding long request timeouts since video generation takes 60-180 seconds.replyUrlis optional, webhook URL for job status callbacks. Receives POST requests with job status updates (created,started,completed,failed). The JSON payload shape matches GET /jobs/jobIdresponse.replyRefis optional, custom reference string passed back in webhook callbacks. Useful for tracking jobs on your end.
Captcha
captchaToken,captchaRetry,captchaOrderare optional, mutually exclusive captcha parameters. See Captcha Parameters for details.
Responses
-
Videos generated successfully. All video data is in the
media[]array.{ "jobId": "j1117123456789304152v-u12345-email:john@gmail.com-bot:google-flow", "operations": [], "media": [ { "name": "a1d95d21-75d2-482d-a354-14ef8802ce66", "projectId": "9f63078c-...redacted...", "workflowId": "fa986834-...redacted...", "workflowStepId": "CAE", "mediaMetadata": { "createTime": "2026-11-15T23:14:02.931677Z", "mediaTitle": "A serene mountain landscape at sunset with camera slowly panning right", "mediaBlobSize": "6895556", "mediaStatus": { "mediaGenerationStatus": "MEDIA_GENERATION_STATUS_SUCCESSFUL" }, "visibility": "PRIVATE", "requestData": { "videoGenerationRequestData": { "backend": "flow", "statusEnum": 3, "aspectEnum": 2, "videoModelControlInput": { "videoModelName": "veo_3_1_t2v_fast", "videoGenerationMode": "VIDEO_GENERATION_MODE_TEXT_TO_VIDEO", "videoAspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE", "videoResolution": "VIDEO_RESOLUTION_720P" } }, "promptInputs": [ { "structuredPrompt": { "parts": [ { "text": "A serene mountain landscape at sunset with camera slowly panning right" } ] } } ], "clientPlatform": "CLIENT_PLATFORM_WEB" } }, "video": { "generatedVideo": { "seed": 123456, "prompt": "A serene mountain landscape at sunset with camera slowly panning right", "model": "veo_3_1_t2v_fast", "baseImageMediaGenerationId": "", "isLooped": false, "aspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE" }, "dimensions": { "length": "8s" }, "operation": { "name": "a1d95d21-75d2-482d-a354-14ef8802ce66" } }, "videoUrl": "https://flow-content.google/video/a1d95d21-...redacted...?Expires=...", "thumbnailUrl": "https://flow-content.google/image/a1d95d21-...redacted...?Expires=...", "mediaGenerationId": "user:12345-email:6a6f...-video:a1d95d21-75d2-482d-a354-14ef8802ce66" } ], "remainingCredits": 18760, "captcha": { "service": "AntiCaptcha", "taskId": "931846120", "durationMs": 15243, "attempts": [ { "service": "AntiCaptcha", "success": true, "durationMs": 15243, "cost": "0.00200", "ip": "203.0.113.24", "createTime": 1794787642, "endTime": 1794787657, "solveCount": 0, "taskId": "931846120", "userAgent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/153.0.0.0 Safari/537.36" } ] } } -
Job created in async mode (
async: true). Video generation is processing in the background.Use GET /jobs/
jobIdto poll for completion status.{ "jobid": "j1117123456789304152v-u12345-email:john@gmail.com-bot:google-flow", "type": "video", "status": "created", "created": "2025-11-17T12:34:56.789Z", "request": { "email": "john@gmail.com", "prompt": "A serene mountain landscape at sunset with camera slowly panning right", "model": "veo-3.1-fast", "aspectRatio": "landscape", "duration": 8, "count": 2, "async": true, "replyUrl": "https://your-domain.com/webhook", "replyRef": "custom-reference-123" }, "response": { "operations": [], "media": [ { "name": "1450903d-...redacted...-9c86f0", "projectId": "9f63078c-...redacted...", "workflowId": "b3e1c0de-...redacted...", "workflowStepId": "CAE", "mediaMetadata": { "createTime": "2025-11-17T12:34:58.102934Z", "mediaTitle": "A serene mountain landscape at sunset with camera slowly panning right", "mediaStatus": { "mediaGenerationStatus": "MEDIA_GENERATION_STATUS_PENDING" }, "visibility": "PRIVATE", "requestData": { "videoGenerationRequestData": { "backend": "flow", "statusEnum": 6, "aspectEnum": 2, "videoModelControlInput": { "videoModelName": "veo_3_1_t2v_fast", "videoGenerationMode": "VIDEO_GENERATION_MODE_TEXT_TO_VIDEO", "videoAspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE", "videoResolution": "VIDEO_RESOLUTION_720P" } }, "promptInputs": [ { "structuredPrompt": { "parts": [{ "text": "A serene mountain landscape at sunset with camera slowly panning right" }] } } ], "clientPlatform": "CLIENT_PLATFORM_WEB" } }, "video": { "generatedVideo": { "seed": 527083, "prompt": "A serene mountain landscape at sunset with camera slowly panning right", "model": "veo_3_1_t2v_fast", "baseImageMediaGenerationId": "", "isLooped": false, "aspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE" }, "dimensions": { "length": "8s" }, "operation": { "name": "1450903d-...redacted...-9c86f0" } } }, { "name": "f2eec9bd-...redacted...-e16f7a", "projectId": "9f63078c-...redacted...", "workflowId": "c47a90b1-...redacted...", "workflowStepId": "CAE", "mediaMetadata": { "createTime": "2025-11-17T12:34:58.102934Z", "mediaTitle": "A serene mountain landscape at sunset with camera slowly panning right", "mediaStatus": { "mediaGenerationStatus": "MEDIA_GENERATION_STATUS_PENDING" }, "visibility": "PRIVATE", "requestData": { "videoGenerationRequestData": { "backend": "flow", "statusEnum": 6, "aspectEnum": 2, "videoModelControlInput": { "videoModelName": "veo_3_1_t2v_fast", "videoGenerationMode": "VIDEO_GENERATION_MODE_TEXT_TO_VIDEO", "videoAspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE", "videoResolution": "VIDEO_RESOLUTION_720P" } }, "promptInputs": [ { "structuredPrompt": { "parts": [{ "text": "A serene mountain landscape at sunset with camera slowly panning right" }] } } ], "clientPlatform": "CLIENT_PLATFORM_WEB" } }, "video": { "generatedVideo": { "seed": 81946, "prompt": "A serene mountain landscape at sunset with camera slowly panning right", "model": "veo_3_1_t2v_fast", "baseImageMediaGenerationId": "", "isLooped": false, "aspectRatio": "VIDEO_ASPECT_RATIO_LANDSCAPE" }, "dimensions": { "length": "8s" }, "operation": { "name": "f2eec9bd-...redacted...-e16f7a" } } } ], "captcha": { "service": "CapSolver", "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e", "durationMs": 5357, "attempts": [ { "service": "CapSolver", "success": true, "durationMs": 5357, "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e", "userAgent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/153.0.0.0 Safari/537.36", "secChUa": "\"Google Chrome\";v=\"153\", \"Not_A Brand\";v=\"8\", \"Chromium\";v=\"153\"" } ] }, "captchaAttemptsCount": 1, "captchaProviders": "CapSolver", "tier": "PAYGATE_TIER_TWO", "sku": "G1_TIER2", "remainingCredits": 24235 } } -
Invalid request (validation error, mode conflict, email mismatch, or content policy violation).
Validation error:
{ "error": "Model 'veo-3.1-quality' does not support R2V mode (MULTI_REFERENCE_NO_STYLE)", "code": 400 }Model not offered on this account (
veo-3.1-lite-low-priorityon an invited member of an Ultra $199 family plan, answered before any captcha is solved):{ "error": "Google no longer offers veo-3.1-lite-low-priority on this account (user@gmail.com). Since 2026-09-23 invited members of a Google AI Ultra $199 family plan do not have it — only the family manager's account keeps it. Use an account that still lists it at flow.google.com, or another model.", "code": 400 }Content policy error:
{ "jobId": "j1117123456789304152v-u12345-email:john@gmail.com-bot:google-flow", "error": "PUBLIC_ERROR_UNSAFE_GENERATION (HTTP 400)", "code": 400, "response": { "error": { "code": 400, "message": "flow.google.com YhhmEf failed", "status": "INVALID_ARGUMENT", "details": [ { "@type": "type.googleapis.com/google.rpc.ErrorInfo", "reason": "PUBLIC_ERROR_UNSAFE_GENERATION" } ] }, "captcha": { "service": "CapSolver", "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e", "durationMs": 5357, "attempts": [ { "service": "CapSolver", "success": true, "durationMs": 5357, "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e" } ] } } }All operations failed (during polling):
The top-level
erroris always the same sentence, so the reason is inresponse.failureReasons— see Why a generation failed above. Each video inresponse.mediaalso carries its own reason (shortened here to the fields that matter).{ "jobId": "j1117123456789304152v-u12345-email:john@gmail.com-bot:google-flow", "error": "All operations failed", "code": 400, "response": { "operations": [], "media": [ { "name": "a629a9...redacted", "projectId": "9f63078c-...redacted...", "mediaMetadata": { "mediaStatus": { "mediaGenerationStatus": "MEDIA_GENERATION_STATUS_FAILED", "error": { "code": 3, "message": "PUBLIC_ERROR_MINOR" } }, "requestData": { "videoGenerationRequestData": { "backend": "flow", "statusEnum": 4, "aspectEnum": 2 } } }, "video": { "operation": { "name": "a629a9...redacted" } }, "mediaGenerationId": "user:12345-email:6a6f...-video:a629a9...redacted" } ], "remainingCredits": 18760, "failureReasons": [ "PUBLIC_ERROR_MINOR" ], "captcha": { "service": "CapSolver", "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e", "durationMs": 5357, "attempts": [ { "service": "CapSolver", "success": true, "durationMs": 5357, "taskId": "14af1dbb-885c-4e25-8121-7a79489dfd0e" } ] } } } -
Invalid API token.
{ "error": "useapi.net ⁝ Unauthorized", "code": 401 } -
Subscription expired or insufficient credits.
{ "error": "useapi.net ⁝ Your account has no subscription, please subscribe to our services before submitting requests", "code": 402 } -
Google rejected the captcha on every try. We retry with a fresh captcha as many times as you allow (
captchaRetry, default 5, or the providers incaptchaOrder) and return this403only when all of them were rejected. The top-levelerrorstarts withcaptcha_quality:, for example"captcha_quality: PUBLIC_ERROR_UNUSUAL_ACTIVITY after 5 attempts. Google refused all 5 captcha tokens for user@example.com, so nothing was generated and no Google Flow credits were used (5 captcha solves were spent). Google is flagging this account's requests as unusual activity. If it keeps happening, use another account, or leave out email so the API picks one. See GET /accounts/captcha-stats summary.by_account". A429withPUBLIC_ERROR_UNUSUAL_ACTIVITY_TOO_MUCH_TRAFFICreads the same way.Two
403s are not a captcha problem, and we answer them after one try instead of retrying.PUBLIC_ERROR_MODEL_ACCESS_DENIEDin the response means Google refused the model on that account, so raisingcaptchaRetryor adding providers will not help. Check that the account’s plan includes the model. Since 2026-09-23 Google no longer offersveo-3.1-lite-low-priorityto invited members of an Ultra $199 family plan — only the family manager’s account keeps it.PUBLIC_ERROR_USER_MINORmeans Google treats the Google account as belonging to someone under 18, and Google Flow is not available to such accounts. Theerrorsays so in plain words. No captcha or retry changes it: use another Google account, or have this one’s age verified with Google first. Withemailomitted, the load balancer sets such an account aside for 1 hour, then tries it again.Nothing was created when this happens. Google checks the captcha before it starts generating, so no Google credits were used and you can resubmit straight away. Your captcha provider still bills each solve.
The usual cause is the Google account, not the captcha provider. Google gives each account its own trust level. On a trusted account it accepts about 9 in 10 captchas, on a distrusted one about 3 in 10, whichever provider solved them. Most customers have a few distrusted accounts among many good ones.
To see which accounts are affected, use GET /accounts/captcha-stats. Its
summary.by_accountshows, for each of your Google accounts, how many captchas were tried and what share Google accepted. GET account/stats withbot=google-flowalso shows how many requests each account finished with each response code.- If one or a few accounts are far below the others, the API already sends them less work: when you leave out
email, each captcha Google rejected on an account in the last 15 minutes lowers its chance of being picked, for generations and for image uploads. They still get some work, so an account that recovers gets its full share back. If you pin requests withemail, move them to your other accounts yourself. Replace an account that stays low for more than a day or two. - If all your accounts drop at the same time, it’s the captcha provider. Configure at least two providers with POST /accounts/captcha-providers, and the API will start with whichever one Google accepts best.
A higher
captchaRetrygets more requests through on a distrusted account, but every extra try is another paid solve. - If one or a few accounts are far below the others, the API already sends them less work: when you leave out
-
Account not found or not configured.
{ "error": "Google Flow account john@gmail.com not found", "code": 404 } -
Video generation polling timeout (after 10 minutes).
{ "error": "Video generation polling timeout after 120 attempts (600s)", "code": 408 } -
Google returns several
RESOURCE_EXHAUSTED429 variants. Check thereasonfield and respond accordingly:Reason Meaning Scope What to do Cooldown PUBLIC_ERROR_USER_REQUESTS_THROTTLEDGoogle is throttling the account (the token was fine). Uploads and generation are metered separately All models on the account — or uploads only, when POST /assets triggered it Run fewer parallel requests, or switch accounts. From POST /assets, retrying the same one rarely clears it ~30 min. From POST /assets, treat it as a minimum — it can persist PUBLIC_ERROR_USER_QUOTA_REACHEDAccount’s overall Flow quota cap All models on the account Add accounts, or omit emailto load-balance~30 min PUBLIC_ERROR_PER_MODEL_DAILY_QUOTA_REACHEDPer-model daily quota Only that model, on that account Switch model, or wait Next UTC midnight PUBLIC_ERROR_UNUSUAL_ACTIVITY_TOO_MUCH_TRAFFICGoogle rejected the captcha on every try (we already retried). Usually a distrusted Google account, sometimes the provider Per request Check which accounts are affected in captcha-stats. Rest or replace those accounts. Add a second provider if all accounts drop together 60s Every
429returns the cooldown two ways — an HTTPRetry-After: <seconds>header and a bodyretryAfter: <ISO timestamp>field. Honor it before retrying.When
emailis omitted, the first three reasons quarantine the account and the load balancer routes around it. If every account is quarantined for the model you get429witherror: "no_eligible_account". Routing details: GET /jobs › Load Balancing Algorithm.Captcha-quality failures prefix the top-level
errorwithcaptcha_quality:(e.g."captcha_quality: PUBLIC_ERROR_UNUSUAL_ACTIVITY_TOO_MUCH_TRAFFIC after 3 attempts") so you can detect them without parsing the nested body. -
503has two distinct causes — check the response body to tell them apart.A transient Google-side outage. Wait 5-10 seconds and retry.
{ "error": { "code": 503, "message": "Service temporarily unavailable.", "status": "UNAVAILABLE" } }A captcha-provider failure — the top-level
erroris prefixedCaptcha service failed:. The problem is your captcha provider (check its API key and balance), not Google. Don’t retry until the provider is back.{ "error": "Captcha service failed: ERROR_ZERO_BALANCE", "code": 503 } -
596 Session Error
Google rejected the account’s session, and the API is re-verifying the account on its own: retry in about a minute. If Google signed the account out, the account keeps answering
596and you get a reconnect email: add it again by following Setup Google Flow.{ "error": "Google Flow account john@gmail.com Google rejected this account session (HTTP 401). We are automatically re-verifying it — please retry in about a minute. If the account was signed out, you will receive a reconnect email.", "code": 596 }
Model
One tab per answer: the finished videos (200, sync), the job record (201, async) and a failed job.
The answer for each status, field by field, and the types it uses (FlowVideoMedia, FlowCaptcha, FlowReference)
-
Video generation completed. Returns full video data with signed URLs.
FlowVideoMediaandFlowCaptchaare described below the tabs.{ jobId: string // the job id, for GET /jobs/{jobid} operations: [] // always empty, kept for older clients: read media[] media: FlowVideoMedia[] // one item per video remainingCredits?: number // the account's Google Flow credits left, as Google last reported them captcha: FlowCaptcha // the captcha solved for this request } -
Job created and processing in background. Structure matches GET /jobs/
jobIdresponse. References you sent come back decoded, asFlowReferenceobjects (described below the tabs).{ jobid: string // the job id, for GET /jobs/{jobid} type: 'video' // the job type status: 'created' // the job status created: string // ISO 8601 time the job was created request: { // the request as you sent it, after validation email: string // the account the job runs on (the one you named, or the one the API picked) prompt: string // the prompt model?: string // the model you asked for aspectRatio?: string // the aspect ratio you asked for duration?: number // the length you asked for, in seconds resolution?: string // the resolution you asked for count?: number // how many videos you asked for startImage?: FlowReference // the first-frame image endImage?: FlowReference // the last-frame image referenceImage_1?: FlowReference // a reference image referenceImage_2?: FlowReference // a reference image referenceImage_3?: FlowReference // a reference image referenceImage_4?: FlowReference // a reference image referenceImage_5?: FlowReference // a reference image referenceImage_6?: FlowReference // a reference image referenceImage_7?: FlowReference // a reference image referenceImage_8?: FlowReference // a reference image referenceImage_9?: FlowReference // a reference image referenceImage_10?: FlowReference // a reference image character_1?: FlowReference // a character character_2?: FlowReference // a character character_3?: FlowReference // a character character_4?: FlowReference // a character character_5?: FlowReference // a character character_6?: FlowReference // a character character_7?: FlowReference // a character character_8?: FlowReference // a character character_9?: FlowReference // a character character_10?: FlowReference // a character referenceAudio_1?: string // a voice: a built-in voice name (lowercase) or your voice reference referenceAudio_2?: string // a voice: a built-in voice name (lowercase) or your voice reference referenceAudio_3?: string // a voice: a built-in voice name (lowercase) or your voice reference referenceAudio_4?: string // a voice: a built-in voice name (lowercase) or your voice reference referenceAudio_5?: string // a voice: a built-in voice name (lowercase) or your voice reference referenceVideo_1?: FlowReference // the video to edit startFrameIndex_1?: number // the first frame of the video to edit endFrameIndex_1?: number // the last frame of the video to edit returnSilentVideos?: boolean // true: a video whose sound Google's filter rejects comes back silent captchaToken?: string // the captcha token you sent captchaRetry?: number // the captcha retries you asked for captchaOrder?: string[] // the captcha providers you listed, in order async: true // answer at once, the job runs in the background replyUrl?: string // your callback URL replyRef?: string // your reference, sent back in callbacks } model?: string // the model that ran, only when it is not the one you asked for (a retired model, routed to its replacement) response: { // Google's answer so far operations: [] // always empty, kept for older clients: read media[] media: FlowVideoMedia[] // one item per video, still being made (no mediaGenerationId or URLs yet) captcha: FlowCaptcha // the captcha solved for this request captchaAttemptsCount: number // how many captcha solves were tried, one per entry of captcha.attempts captchaProviders: string // the providers tried, comma-separated, in the order tried tier: string // the account's Google plan tier when the job started, as Google reports it, e.g. "PAYGATE_TIER_TWO" sku: string // the account's Google plan code when the job started, e.g. "G1_TIER2" remainingCredits?: number // the account's Google Flow credits left, as Google reported them at submit } } -
Error response structure (applies to both sync and async modes).
{ jobId?: string // the job id, when the job had been created error: string // what went wrong, e.g. "All operations failed" or "PUBLIC_ERROR_UNSAFE_GENERATION (HTTP 400)" code?: number // the HTTP status retryAfter?: string // ISO 8601 time to retry after, on a 429 (the Retry-After header carries the same in seconds) response?: { // what the job held when it failed failureReasons?: string[] // why every video failed: Google's terminal codes and classifier labels, de-duplicated, // e.g. PUBLIC_ERROR_IP_INPUT_IMAGE or IP_PROHIBITED. Present only when all videos // failed and Google gave a reason operations?: [] // always empty, kept for older clients media?: FlowVideoMedia[] // every video of the job, when none was made: each carries its own reason in mediaMetadata.mediaStatus remainingCredits?: number // the account's Google Flow credits left, as Google last reported them error?: string | { // Google's error when Google refused the request, or our own text (e.g. a polling timeout) code: number // HTTP status Google answered message: string // the error message, e.g. "flow.google.com YhhmEf failed" status: string // Google's status name, e.g. "RESOURCE_EXHAUSTED" details: Array<{ // Google's error details '@type': string // the type name of the detail reason: string // Google's reason code, e.g. "PUBLIC_ERROR_USER_QUOTA_REACHED" }> } captcha?: FlowCaptcha // the captcha tried or used captchaAttemptsCount?: number // how many captcha solves were tried, when the job failed before its first poll captchaProviders?: string // the providers tried, comma-separated, when the job failed before its first poll tier?: string // the account's Google plan tier, when the job failed before its first poll sku?: string // the account's Google plan code, when the job failed before its first poll } message?: string // on the load-balancer 429 only (error no_eligible_account: every account is quarantined for this model): the explanation skipReasons?: Array<{ // on the load-balancer 429 only: why each account was left out email: string // the account left out reason: string // Google reason that triggered the quarantine (a 429 reason, or the MODEL_ACCESS_DENIED or USER_MINOR 403) model: string // Quarantined model, or "*" for account-wide }> }
The types the tabs refer to:
FlowVideoMedia: each item of media in the tabs above.
{
name: string // Google's media id of the video
projectId: string // the Google Flow project the video is in
workflowId?: string // the Google Flow workflow (project item) that holds the video
workflowStepId?: string // the workflow step, e.g. "CAE" (left out when Google's record has none)
mediaMetadata: { // Google's record of the video
createTime?: string // ISO 8601 time Google created the video
mediaTitle?: string // the prompt, as Google Flow titles the video (an upscale has none)
mediaBlobSize?: string // the MP4's size in bytes, as a string, once the video is done
mediaStatus: { // whether the video was made
mediaGenerationStatus: string // MEDIA_GENERATION_STATUS_PENDING | MEDIA_GENERATION_STATUS_ACTIVE |
// MEDIA_GENERATION_STATUS_SUCCESSFUL | MEDIA_GENERATION_STATUS_FAILED |
// MEDIA_GENERATION_STATUS_UNKNOWN
error?: { // why the video failed
code: number // Google's error code
message: string // Google's reason, e.g. "PUBLIC_ERROR_IP_INPUT_IMAGE"
}
failureReasons?: string[] // Google's classifier labels for a failed video, e.g. "IP_PROHIBITED"
}
visibility?: string // "PRIVATE" (left out on a failed video)
requestData: { // what the video was made from, as Google echoes it
videoGenerationRequestData: { // the generation settings
backend: 'flow' // the video was made on `flow.google.com`
statusEnum: number | null // Google's raw status number: 6 queued, 1 and 2 running, 3 done, 4, 5 and 7 failed
aspectEnum: number | null // Google's raw aspect number: 2 landscape, 1 portrait
videoModelControlInput?: { // the model settings Google ran
videoModelName?: string // Google's own key for the model that ran
videoGenerationMode?: string // VIDEO_GENERATION_MODE_TEXT_TO_VIDEO | VIDEO_GENERATION_MODE_REFERENCE_TO_VIDEO |
// VIDEO_GENERATION_MODE_VIDEO_EXTENSION (left out for the other modes)
videoAspectRatio?: string // VIDEO_ASPECT_RATIO_LANDSCAPE | VIDEO_ASPECT_RATIO_PORTRAIT
videoResolution?: string // VIDEO_RESOLUTION_360P | VIDEO_RESOLUTION_720P | VIDEO_RESOLUTION_1080P |
// VIDEO_RESOLUTION_4K
}
}
promptInputs?: Array<{ // the prompt as Google received it (an upscale has none)
structuredPrompt: { // the prompt in parts
parts: Array<{ // a run of text, or an @-mention
text?: string // a run of prompt text
reference?: object // an @-mention: { media: { mediaId, handle } }, { audio: { mediaId, handle } } or { entity: { entityId, handle } }
}>
}
}>
clientPlatform?: string // "CLIENT_PLATFORM_WEB"
}
}
video: { // the video's details
generatedVideo?: { // how Google made it (left out until Google reports it)
seed: number // the seed Google used (0 when it reports none)
prompt: string // the prompt as Google ran it (Veo rewrites it), an empty string on an upscale
model: string // Google's own key for the model that ran
baseImageMediaGenerationId?: string // an empty string on every video so far
isLooped: false // always false
aspectRatio: string // VIDEO_ASPECT_RATIO_LANDSCAPE | VIDEO_ASPECT_RATIO_PORTRAIT
// (an empty string when Google reports none)
upsampleMetadata?: { // on an upscaled video
videoUpsampleResolution: string // VIDEO_UPSAMPLE_RESOLUTION_720P | VIDEO_UPSAMPLE_RESOLUTION_1080P
// (left out for 4K)
}
}
dimensions?: { // the clip's measurements (an upscaled clip has none)
length: string // the clip's length, e.g. "8s"
}
operation: { // the generation operation
name: string // the same as name
}
}
mediaGenerationId?: string // the id to pass to the other endpoints (upscale, extend, GIF, concatenate, assets), on every finished answer, left out of the 201 answer
videoUrl?: string // signed MP4 download URL, valid about 6 hours, once the video is done
thumbnailUrl?: string // signed JPEG thumbnail URL, valid about 6 hours
}
FlowCaptcha: the captcha in the tabs above.
{
service?: string // the provider whose token Google accepted: `CapSolver`, `AntiCaptcha`,
// `YesCaptcha`, `CapMonster`, `SolveCaptcha`, `2Captcha`, `EzCaptcha` or
// `UserProvided`. Always on a success answer
taskId?: string // that provider's task id (none for `UserProvided`)
durationMs?: number // ms from the first solve to the token used. Always on a success answer
error?: string // why no captcha token could be had, on a failed request
attempts: Array<{ // every solve tried, in order, failed ones included
service: string // the provider asked
taskId?: string // that provider's task id
durationMs: number // how long this solve took, in ms
success: boolean // true when the provider returned a token (Google may still have refused it)
error?: string // why the solve failed
userAgent?: string // the browser user agent the provider solved with, when it reports one
secChUa?: string // the Sec-CH-UA value the provider solved with, when it reports one (may be empty)
cost?: string // the provider's price for this solve, as the provider reports it
ip?: string // the IP address the provider reports for this solve
createTime?: number // Unix time the provider started the solve, as it reports it
endTime?: number // Unix time the provider finished the solve, as it reports it
solveCount?: number // the provider's own solve counter for this task
}>
}
FlowReference: each image, character or video reference in request above.
{
user: number // your useapi.net user id, read from the reference
email: string // the Google account the media belongs to, read from the reference
type: 'image' | 'video' | 'character' // what the reference points to
id: string // Google's media id (image or video), or the character's entity id
referenceId: string // the reference you sent
meta?: { // characters only
imgs: number // how many images the character holds (they count toward the model's image limit)
voiceWorkflowId?: string // the voice attached to the character, when it has one
}
}
Missing media URLs
Media URL fields are optional and can be absent even when generation succeeded. When videoUrl, thumbnailUrl, fifeUrl, servingBaseUri, previewUrl or audioUrl is missing, the media itself is fine — only the download link could not be produced at that moment.
This is rare. Across the seven days of request logs we keep, it happened on one day only — twice, for about four hours and about forty minutes — and not at all on the other six. Google issues those links from an endpoint that limits how often it will answer a given network address, so it is Google limiting our outbound calls rather than useapi.net limiting you, and nothing you do triggers it. While it lasts it affects every account generating at that moment, then clears on its own. Your generation still completed and was charged as usual.
On a media item that generated successfully — mediaStatus.mediaGenerationStatus is MEDIA_GENERATION_STATUS_SUCCESSFUL — a missing URL means “not yet”, not “failed”, so read the field before using it and treat its absence as a retry condition. A media item that failed generation never carries a URL, and retrying will not produce one.
const url = media.videoUrl;
if (!url) {
// Not an error — the link is not available yet. See the options below.
}
How to get the link
Which route to use depends on whether you can wait.
If you are already polling the job, do nothing special. Completed video jobs re-resolve missing URLs automatically on GET /jobs/jobId, and the recovered link is saved, so a later poll returns it with no extra work on your side. This applies to video jobs created within the last 24 hours. Once the job is completed and a URL is still missing, leave at least a minute between polls — a job re-resolves at most once a minute, and the block clears only once traffic stops, so a tight retry loop lengthens the wait rather than shortening it.
If you hold a mediaGenerationId but no longer have the job, ask for that one media again with GET /assets/mediaGenerationId. That endpoint says which case you are in rather than leaving you to guess. A 503 with a Retry-After header means the link is coming and tells you how long to wait, so wait that long and ask again. A 404 means Google has no such media, and no amount of retrying will change it.
If you cannot wait, download the video through us with GET /assets/mediaGenerationId?raw=true. It reads the media over a different Google route that the block does not touch, so it works immediately and keeps working while the other two return nothing. It is video only.
Prefer the two link-based routes when you can. They hand you a link and you download straight from Google. The raw form moves the whole file through useapi.net on every call, which makes it a way out of a stuck download rather than your normal download path.
Characters and voices resolve their URLs as they are read, so simply requesting GET /characters/ref or GET /voices/ref again is enough. Wait a minute before retrying — a response that came back without its URL is held briefly, so an immediate repeat returns the same one.
None of this applies to images. Whenever an image URL is missing, POST /images puts the image itself in the response instead, base64-encoded in encodedImage — so you already have the picture and there is nothing to retry.
Examples
-
curl -X POST \ -H "Authorization: Bearer YOUR_API_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "prompt": "A serene mountain landscape at sunset with camera slowly panning right", "model": "veo-3.1-fast", "aspectRatio": "landscape", "duration": 8, "count": 2 }' \ "https://api.useapi.net/v1/google-flow/videos" > response.json # Extract video URLs using jq cat response.json | jq -r '.media[0].videoUrl' cat response.json | jq -r '.media[1].videoUrl' # Download videos using curl curl "$(cat response.json | jq -r '.media[0].videoUrl')" --output video_1.mp4 curl "$(cat response.json | jq -r '.media[1].videoUrl')" --output video_2.mp4 -
const token = 'YOUR_API_TOKEN'; const apiUrl = 'https://api.useapi.net/v1/google-flow/videos'; const response = await fetch(apiUrl, { method: 'POST', headers: { 'Authorization': `Bearer ${token}`, 'Content-Type': 'application/json' }, body: JSON.stringify({ prompt: 'A serene mountain landscape at sunset with camera slowly panning right', model: 'veo-3.1-fast', aspectRatio: 'landscape', duration: 8, count: 2 }) }); const result = await response.json(); console.log('Generated videos:', result.media.length); // Download videos for (const [index, item] of result.media.entries()) { console.log(`Video ${index + 1} seed:`, item.video?.generatedVideo?.seed); console.log(`mediaGenerationId:`, item.mediaGenerationId); console.log(`Video URL:`, item.videoUrl); // Download video (Node.js) const videoResponse = await fetch(item.videoUrl); const videoBuffer = await videoResponse.arrayBuffer(); const fs = require('fs'); fs.writeFileSync(`generated_video_${index + 1}.mp4`, Buffer.from(videoBuffer)); } -
import requests token = 'YOUR_API_TOKEN' api_url = 'https://api.useapi.net/v1/google-flow/videos' headers = { 'Authorization': f'Bearer {token}', 'Content-Type': 'application/json' } data = { 'prompt': 'A serene mountain landscape at sunset with camera slowly panning right', 'model': 'veo-3.1-fast', 'aspectRatio': 'landscape', 'duration': 8, 'count': 2 } response = requests.post(api_url, headers=headers, json=data) result = response.json() print(f"Generated {len(result['media'])} videos") # Download videos for index, item in enumerate(result['media']): video = item.get('video', {}).get('generatedVideo', {}) print(f"Video {index + 1} seed:", video.get('seed')) print(f"mediaGenerationId:", item.get('mediaGenerationId')) # Download video video_response = requests.get(item['videoUrl']) with open(f'generated_video_{index + 1}.mp4', 'wb') as f: f.write(video_response.content)