Short answer: 8frame runs up to 10 generations at once per account, on every plan. The canvas takes images in PNG, JPG, WebP, GIF, AVIF and BMP up to 50 MB (not HEIC), video in MP4, WebM and MOV, and audio in MP3 and WAV. Prompt limits are set per model, from 800 characters on Wan 3.0 to 8,000 on Grok Imagine Image. One clip runs from 1 second up to 30 (Wan 3.0 and Seedance 2.5). Video tops out at 2K: no video model outputs 4K and there is no video upscaler. Inviting collaborators takes a Creator plan (up to 3 members) or Agency (up to 30). Below, each limit gets a one-line answer and a link to the page with the detail, all read from 8frame's code and pricing page on 2026-10-01.
TL;DR
- 10 generations at once, counting the ones waiting to start and the ones running, on Starter, Creator and Agency alike
- Image uploads: PNG, JPG/JPEG, WebP, GIF, AVIF, BMP, up to 50 MB. HEIC is not accepted on the canvas
- Prompts: 800 characters (Wan 3.0) to 8,000 (Grok Imagine Image), set per model
- Clips: 1 to 30 seconds in one generation; 30 only on Wan 3.0 and Seedance 2.5
- Video resolution: 2K at most (Hailuo H3, Creatify Boreal). No 4K, no video upscaler
- Collaborators: none on Starter, up to 3 on Creator, up to 30 on Agency
- Credits roll over while you are subscribed, are charged to whoever runs the generation, and come back automatically when a generation fails
- MCP connector: waits up to 10 minutes for a result, then hands back a task ID
Every limit, one table
| Limit | Value | Where it applies | Detail |
|---|---|---|---|
| Generations at once | 10 active (waiting or running) | Every account, every plan; connector generations included | This page, below |
| Image upload | PNG, JPG/JPEG, WebP, GIF, AVIF, BMP, up to 50 MB; no HEIC | Canvas uploads | This page, below |
| Video and audio upload | Video MP4, WebM, MOV; audio MP3, WAV | Canvas uploads; clip length and size depend on the model | Edit an existing video with AI |
| Image size for upscaling | Up to 6,144 x 6,144 px; a larger image is scaled down, not refused | Topaz, Clarity, BRIA | AI image resolution by model |
| Prompt length | 800 to 8,000 characters | Per model: Wan 3.0 lowest, Grok Imagine Image highest | Prompt length limits by model |
| Clip length | 1 to 30 s per generation | Per model; 30 s on Wan 3.0 and Seedance 2.5 | Clip length limits by model |
| Clips per merge | 2 to 10, each cut to a maximum of 1 to 15 s (default 5) | Montage node | How to use Merge Videos |
| Reference images | Up to 14 on image models, up to 10 on video models; none on Wan 3.0 and Kling v3 | Per model | Reference image limits by model |
| Video resolution | 2K at most; no 4K, no video upscaler | Every video model | AI video resolution by model |
| Generated image resolution | 4K at most; the BRIA upscaler goes to 8,192 x 8,192 px | Nano Banana 2 and Nano Banana Pro | AI image resolution by model |
| Images per run | 1 on most models; 1 to 4, 1 to 6, or always 4 | Nano Banana 2 and GPT Image 1 to 4; Qwen Image 3 1 to 6; Midjourney 4 | Images per run by model |
| Collaborators | Starter none; Creator up to 3; Agency up to 30 | Per board, by the board owner's plan | Collaborative AI canvas for teams |
| Credit rollover | Unused credits carry over | While subscribed | AI video credits explained |
| Failed generations | Refunded automatically, in full | Every generation | Do failed AI generations cost credits |
| Credit sharing | Not pooled: charged to whoever runs the generation | Shared boards | This page, below |
| Connector wait | Up to 10 minutes, then a task ID | MCP connector | Generate AI video from Claude |
| Connector daily spend | A per-user daily credit cap (we do not publish the number) | MCP connector only | Generate AI video from Claude |
Generations at once: 10, on every plan
8frame counts your active generations: the ones waiting to start and the ones running. At 10, the next one is refused before any credits are taken, with the message "Limit reached: 10 active tasks, please wait for some to finish." As soon as one finishes or fails, its slot frees up.
The limit is the same on Starter, Creator and Agency, and a bigger plan does not buy a faster queue either: 8frame orders the queue by tool type, not by plan. Generations you start through the MCP connector count toward the same 10.
Slow models hold a slot longest. A 5-second Seedance 2.5 clip at 720p has a median of 287 seconds on 8frame, so ten of them block everything else for close to five minutes. If you also want quick image drafts, start fewer long clips at once. Median times for other models are in how long AI video generation takes.
Uploads: formats and sizes
- Images: PNG, JPG/JPEG, WebP, GIF, AVIF and BMP, up to 50 MB per file.
- Video: MP4, WebM and MOV.
- Audio: MP3 and WAV.
- HEIC, the iPhone photo format, is not accepted on the canvas. Export the photo as JPG first, or use the MCP upload page, which takes JPG, PNG, WebP, GIF and HEIC and converts HEIC to JPEG.
Image dimensions. The image upscalers (Topaz, Clarity and BRIA) take images up to 6,144 x 6,144 pixels. A larger image is not refused: 8frame scales it down to fit before the run, and it does the same whenever a model's input limit is smaller than your upload. What does stop a run is an image below a model's minimum size, so very small crops are the ones to watch.
How long and how large an uploaded clip can be depends on the model you feed it to. The video-in routes and their input lengths are in edit an existing video with AI.
Prompt length: 800 to 8,000 characters
The limit follows the model, in characters, spaces included. Wan 3.0 and Wan 3.0 Prime take the fewest, 800. Seedance 2.5 takes 2,000, half of Seedance 2.0's 4,000. Kling, every version on 8frame, takes 2,500; Grok Imagine video 4,096; Nano Banana 5,000; Grok Imagine Image the most, 8,000. Text to speech takes up to 5,000 characters of script per generation.
The prompt box stops at the model's limit, and a longer paste is cut at it, so recheck the length after switching models. Every model's number is in prompt length limits by model.
Clip length: 1 to 30 seconds per generation
- Longest in one pass: Wan 3.0 (2 to 30 s) and Seedance 2.5 (4 to 30 s).
- Shortest: 1 second, on Grok Imagine (1 to 15 s), Pruna p-video (1 to 20 s) and Creatify Boreal (1 to 20 s).
- Veo 3.1: 4, 6 or 8 seconds; with a start image, 8 only.
- Most other models stop at 10 or 15 seconds.
Length is priced: a 30-second Wan 3.0 clip is 142 credits at 480p and 567 at 1080p, its default. For anything longer than 30 seconds, generate several clips and join them. The Montage node joins 2 to 10 clips for 10 credits per merge, and it cuts each clip to a maximum length you set, 1 to 15 seconds, default 5: raise it before joining longer clips. Lip sync clips follow the length of your audio, up to 35 seconds on Pruna Avatar and OmniHuman 1.5.
Every model's range is in clip length limits by model; joining is covered in how to use Merge Videos.
Reference images: up to 14 on images, up to 10 on video
- Image models: Seedream 14, Nano Banana 10, GPT Image 10, Reve 6, Grok Image 3, Qwen Image 3 up to 3 when editing.
- Video models: Grok Imagine's reference mode 1 to 10, Happy Horse 1 to 9 character references, Seedance 9 on the canvas (plus one reference clip), Gemini Omni Flash 7.
- No reference images: Wan 3.0 and Kling v3 animate one start image, and Kling v3 requires it.
The full list, and which models drop the start image when references are attached, is in reference image limits by model.
Resolution: 2K for video, 4K for images
- Video: 2K at most, from Hailuo H3 (always 2K) and Creatify Boreal (when you choose 2K). Veo 3.1 and Veo 3.1 Fast deliver 1080p, Veo 3.1 Lite 720p, Seedance 480p or 720p. No video model on 8frame outputs 4K, and there is no video upscaler, so generate at the resolution you will deliver.
- Watch the Wan default: Wan 3.0 starts at 1080p, 95 credits for 5 seconds against 24 at 480p.
- Images: Nano Banana 2 and Nano Banana Pro go to 4K (22 and 42 credits per image). The BRIA upscaler enlarges 2x or 4x, up to 8,192 x 8,192 pixels; Clarity always enlarges 2x.
Details: AI video resolution by model and AI image resolution by model.
Images per run
Most image models return one image per run. Nano Banana 2 and GPT Image let you ask for 1 to 4 and Qwen Image 3 for 1 to 6, each billed per image. Midjourney always returns four for one price, 32 credits on V8, the default. Through the MCP connector, every image tool returns one image per call except Midjourney, which returns four. Every model is in images per run by model.
Collaborators: set by the board owner's plan
- Starter (or no plan): you cannot invite collaborators.
- Creator: up to 3 members on a board.
- Agency: up to 30 members on a board.
The count is per board, checked against the plan of the board's owner, and only the owner can invite. Collaborators need their own 8frame account: ordinary boards have no public share link. More on roles and live editing in collaborative AI canvas for teams.
Credits: rollover, refunds, no pooling
- Plans: Starter $19 a month for 1,000 credits, Creator $49 for 3,000, Agency $499 for 44,000. A credit is about $0.01 at pack rate; top-up packs start at $10 for 1,000. Signing up adds no credits.
- Rollover: unused credits carry over from month to month while your subscription is active.
- Not pooled: each generation is charged to the person who runs it, on that person's own plan. A collaborator who generates on your board spends their credits, not yours.
- Failed generations: refunded automatically and in full. A generation that completed but missed the brief is not a failure and is not refunded. See do failed AI generations cost credits.
MCP connector: 10-minute wait, daily cap
- Wait: the connector holds a request open for up to 10 minutes while the generation runs. If it is still running after that, or your client does not support progress updates, you get a task ID, and the 8frame-check-task tool fetches the result later. Checking a task costs no credits, and the generation keeps running in the meantime.
- Daily cap: credits spent through the connector have a per-user daily cap, a safety limit on what an assistant can spend for you. It resets each day (UTC). We do not publish the number. There is also a short-term limit on bursts of connector requests. Generations on the canvas count toward neither.
Setup and what every connector tool costs: generate AI video from Claude.
FAQ
How many generations can I run at once on 8frame? 10 per account, counting the ones waiting to start and the ones running. The next one is refused with "Limit reached: 10 active tasks" until a slot frees up.
Does a higher plan let me run more generations at once? No. The limit is 10 on Starter, Creator and Agency, and the queue is ordered by tool type, not by plan.
What is the maximum upload size on 8frame? 50 MB for images. For video and audio, the limit depends on the model you use the file with; see edit an existing video with AI.
Can I upload HEIC photos? Not on the canvas: export them as JPG, PNG or WebP first. The MCP upload page accepts HEIC and converts it to JPEG.
What is the longest video 8frame can generate in one go? 30 seconds, on Wan 3.0 or Seedance 2.5. Join clips with the Montage node for anything longer.
Can 8frame generate 4K video? No. The highest video output is 2K, from Hailuo H3 and Creatify Boreal, and there is no video upscaler. Images go to 4K on Nano Banana 2 and Nano Banana Pro.
How many people can I invite to a board? None on Starter, up to 3 on Creator and up to 30 on Agency, counted per board against the owner's plan.
Do unused credits expire? They roll over from month to month while your subscription is active.
Are credits shared between collaborators? No. Each generation is charged to the person who runs it.
Why did the MCP connector give me a task ID instead of a result? The generation was still running after 10 minutes, or your client does not support progress updates. Ask your assistant to check the task: checking costs no credits.
Sources
- Generations at once (10 active tasks, pending and processing, the same for every account; checked before credits are deducted; the error text): 8frame backend
TasksService.checkUserTaskLimitin src/tasks/application/tasks.service.ts, origin/main, read 2026-10-01; 8frame content fact sheet, "Concurrency" entry, read 2026-10-01 - Queue ordered by tool type, no plan-based priority; Seedance 2.5 median 287 s for 5 s at 720p: 8frame content fact sheet, "Generation time by model" entry, read 2026-10-01
- Upload formats, 50 MB image size, HEIC on the canvas and on the MCP upload page: 8frame front src/features/shared/constants/validation.ts, origin/main, read 2026-10-01; 8frame content fact sheet, "Upload formats" entry, read 2026-10-01
- 6,144 x 6,144 px and downscaling instead of refusal: 8frame backend src/tasks/config/tool-registry.ts (input limits of Image Upscale and Topaz), src/tasks/presentation/tasks.controller.ts (general image limits) and src/tasks/services/tool-constraint-validation.service.ts (resize, minimum size enforced), origin/main, read 2026-10-01
- Prompt limits (Wan 3.0 800, Seedance 2.5 2,000, Seedance 2.0 4,000, Kling 2,500, Grok Imagine 4,096, Nano Banana 5,000, Grok Imagine Image 8,000, text to speech 5,000) and the prompt box stopping at the limit: 8frame front src/features/tools/schemas.ts and PromptInput, backend packages/shared/src/constants/tool-validation.ts, origin/main, read 2026-10-01
- Clip lengths, Montage (2 to 10 clips, 1 to 15 s per clip, default 5, 10 credits), lip sync length, resolutions, reference image caps, images per run and prices: 8frame tool registry credit functions and front tool schemas, origin/main, read 2026-10-01; 8frame content fact sheet, sections 2 and 3, read 2026-10-01
- Collaborators (Creator up to 3 members, Agency up to 30): 8frame pricing page copy, read 2026-10-01; counted per board against the owner's plan, owner-only invites: backend
SubscriptionLimitsService.canAddCollaboratorsandWorkflowsService.addCollaborator, origin/main, read 2026-10-01; 8frame content fact sheet, section 5 - Plans, credit value, rollover, no pooling, no credits at signup: 8frame content fact sheet, sections 1 and 5, read 2026-10-01
- MCP connector (10-minute wait, task ID, check-task costs no credits, daily cap on connector spend by UTC day, burst limit): 8frame backend apps/mcp-server/src/client/backend.ts, tools/factory.ts and server.ts, src/tasks/services/mcp-daily-cap.service.ts, origin/main, read 2026-10-01; 8frame content fact sheet, "MCP CONNECTOR" entry, read 2026-10-01
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.