← Back to blog

Upload Limits for AI Video and Image Models in 2026: Size, Resolution, Format

Per-model upload limits on 8frame: minimum image size, maximum file size, accepted formats, and video and audio length caps, plus which limits the canvas fixes for you and which it cannot.

Every model has its own upload limits, and they differ more than people expect: Veo 3.1 takes an 8 MB start image while Seedance takes 50 MB, and Veo needs at least 1280 by 720 pixels while Kling is happy with 300 by 300. On 8frame most of these limits are handled for you. An image that is too large, too many pixels or in the wrong format is resized or converted before it reaches the model. The one thing that cannot be fixed is an image that is too small.

TL;DR

What the canvas accepts

Type Formats Upload size
Images PNG, JPG, WebP, GIF, AVIF, BMP up to 50 MB
Video MP4, WebM, MOV per model, see below
Audio MP3, WAV per model, see below

That is the door. Each model then has its own limits, applied when you connect the file to it. So a 30 MB photo uploads fine, and is then resized for a model that takes 10 MB.

The hard limit: minimum image size

This is the check that actually stops a run. Nothing is upscaled automatically, because enlarging an image quietly would lower the quality you are paying for.

Model Minimum image size
Veo 3.1 1280 px wide and 720 px tall
Video Effect (Higgsfield) 512 x 512
Midjourney V8, V7 (the Artistic tool on the canvas) 512 x 512
Wan 3.0, 3.0 Prime, 2.5, 2.2 360 x 360
Kling (v3, 2.6 Pro, O1, O3) 300 x 300
Happy Horse 300 x 300
FLUX.2 [pro], Flux Edit 256 x 256
Reve 128 x 128

The Veo row needs a note. The check is literal: at least 1280 wide and at least 720 tall. A vertical 720 x 1280 image fails the width check, so a vertical start frame for Veo has to be at least 1280 px wide. Details in Veo 3.1 start image requirements.

A 1024 x 1024 square is too small for Veo 3.1 and fine for everything else in the table.

Maximum file size by model

These are the models' own limits. On 8frame an image over them is resized for you; if you call a model's API directly or use another platform, they are hard limits.

Model Max image file
Veo 3.1 8 MB
Kling, Happy Horse, Hailuo H3, Gemini Omni Flash, Grok Imagine 10 MB
Nano Banana, GPT Image, Reve, Topaz Image Upscale 10 MB
Midjourney V8, V7 (Artistic) 12 MB
Background removal 12 MB
Video Effect (Higgsfield) 15 MB
Runway Gen-4 Turbo 16 MB
Smart Edit 20 MB
Wan 3.0, 3.0 Prime, 2.5, 2.2 25 MB
Seedance 2.5, 2.0 50 MB
Seedream, Flux Edit, Clarity upscaler 100 MB

Kling's numbers match what resellers of its API publish: JPG, JPEG or PNG, at least 300 by 300, at most 10 MB, aspect ratio between 1:2.5 and 2.5:1 (Magnific's Kling 3 API docs, checked 2026-09-28). On 8frame, an image outside that aspect ratio range is adjusted to fit rather than refused.

HEIC photos from an iPhone

The 8frame upload does not take HEIC, the format iPhones save photos in by default. Two ways around it, both from Apple's own guide (checked 2026-09-28):

Video and audio inputs

Unlike images, video and audio are not resized to fit, so these limits are the ones to plan around. Seedance's reference audio is the only input trimmed for you.

Model Input Limit
Gemini Omni Flash Video to edit or extend 10 seconds, 100 MB, MP4 or MOV
Grok Imagine Video 15 seconds, 100 MB
Happy Horse Video to edit 3 to 60 seconds, 100 MB, MP4 or MOV
Kling O1 Video 100 MB, MP4, MOV or WebM
Seedance 2.5 Reference videos up to 10, each up to 30 seconds
Seedance 2.0 Reference videos up to 3, each up to 15 seconds
Seedance 2.5, 2.0 Reference audio MP3 or WAV, up to 12 MB, trimmed for you to 30 s (2.5) or 15 s (2.0)
OmniHuman Driving audio up to 10 MB and 35 seconds
Hailuo H3 lip sync Audio up to 14.8 seconds

One regional limit also applies to Gemini Omni Flash: editing or extending uploaded video is not available in the EEA, Switzerland or the UK, per Google. More in why an AI generation fails.

FAQ

Why does 8frame say my image is too small? The model you picked has a minimum size, and the image is under it. Veo 3.1 needs 1280 by 720; most others need 300 to 512 pixels on each side. Upscale the image or choose a model with a lower minimum.

Do I need to shrink a large photo before uploading? Usually not on 8frame. Images up to 50 MB upload, and anything over a model's limit is resized for it. If a model still reports the file as too large, export it under 10 MB and try again.

Why won't my iPhone photo upload? It is probably HEIC. Set Settings, Camera, Formats to Most Compatible for new photos, or convert the file to JPEG.

What is the maximum video length I can upload for editing? It depends on the model: 10 seconds for Gemini Omni Flash, 15 for Grok Imagine, 60 for Happy Horse.


Check the minimum size first; the canvas handles the rest. The 8frame canvas is free and unlimited, and generation is paid from $19/month.

Related articles

comparisonHow Long Can an AI Video Be? Clip Length Limits by Model in 2026comparisonSora 2 API Shuts Down Sept 24: What to UseguideAI Generation Refusal Messages, Decoded: What Each Error Actually Means

Make it
move.

Stay in the loop

Be the first to hear about our launch and get product updates