Most AI video tools give you one reference image and hope for the best. You upload a character, type a prompt, and get a clip where the face drifts, the outfit shifts, and the proportions change between shots. That is fine for a quick social post. It breaks down the moment you need the same character across a 30-second story.
DomoAI just launched Omni Reference, and it solves exactly that problem. Instead of one image, you can now feed up to 30 images, 10 video clips, and 10 audio files into a single generation. Each reference gets an assigned role. You tell the model which image carries the character’s identity, which clip carries the motion, and which file carries the sound. The model stops guessing and starts listening.
It runs on Seedance 2.5 right now, with MiniMax H3 coming to the same surface soon. The result is a workflow that lets you build character-consistent, multi-scene AI video without leaving one page.
What Omni Reference Actually Does
The headline feature is scale. Up to 50 references in one generation. But the part that matters is the role assignment.
Before this, you would upload a single reference image and cross your fingers. Now you attach a character sheet, a motion clip, a voice file, and first and last frame references, and the model uses all of them with purpose.
You can also run a generation from audio alone. No image required. You upload a song or voiceover and let the model build visuals from the sound. That opens up music video workflows that were impossible in consumer AI video tools until now.
Clip length goes up to 30 seconds. That is double the previous 15-second cap on DomoAI’s Seedance 2.0 features. The trade-off is resolution. Omni Reference runs at 480P or 720P, while the older Seedance 2.0 features go up to 1080P. You are exchanging sharpness for length and control.
What the Output Looks Like
Japanese creator JPN GIRL used Omni Reference to produce an 80-second AI short film assembled from multiple generated shots, all with a consistent character throughout. That is not a demo reel. That is a workflow for recurring characters across scenes.
On Reddit, Seedance 2.5 clips in vlog-style ad formats earned over 100 upvotes, with users pointing specifically to motion coherence on longer clips. On X, creator @techhalla built frozen-time VFX sequences using timecoded shot-list prompts, reaching 191K views on a single post.
The pattern is clear. When you give the model enough references with clear roles, character consistency stops being a happy accident and starts being something you can plan for.
How to Use It
Omni Reference is live now at domoai.app/create/omni-reference. It is not switched on by default, so you need to open the surface and select it.
New accounts start with 30 credits, which covers one generation at 480P and four seconds. That is enough to test the feature once and see whether the output quality meets your standard before committing more.
The credit cost varies based on what you upload, how long the clip runs, and which resolution you pick. There is no flat rate. Configure your references first, then read the exact credit number on the Generate button before you commit.
What to Look Out For
Three things stand out.
First, the role-based reference system is a genuine shift in how AI video tools work. Instead of the model interpreting your single image and hoping it got the vibe right, you assign each input a job. Character identity goes here. Motion style goes there. Audio timing goes in this slot. That is how production pipelines actually think, and it is now available in a consumer tool.
Second, MiniMax H3 is coming to the same surface. When it arrives, Omni Reference becomes a model picker. You will be able to choose between Seedance 2.5 for longer multi-reference sequences and MiniMax H3 for shorter, sharper clips, all without leaving the page or rebuilding your reference set.
Third, the 720P ceiling is real. If you need 1080P output, you still need the older Seedance 2.0 features. Omni Reference is for control and length, not final delivery resolution. Upscaling after generation is the practical play.
What This Unlocks for Creators
The biggest unlock is character consistency across scenes. If you are making an animated series, a motion comic, a product demo with a recurring spokesperson, or a music video where the performer needs to look the same from verse to chorus, Omni Reference gives you a workflow that treats your references as instructions instead of suggestions.
Audio-led generation is the second unlock. Being able to build visuals from a song file means music video production is no longer limited to people who can animate by hand or afford motion capture. You upload the track, set your visual references, and the model builds clips that follow the rhythm.
The third is scene-by-scene assembly. Instead of generating one long clip and hoping the character holds together, you build each shot with the same reference set, then arrange the results around your story structure. That is closer to how real video production works than any single-prompt generation tool has gotten.
Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.
