Characters that stay on model
Headshots or a turnaround sheet lock in face shape, hair, clothing and proportions as the character walks into new shots and settings.
Attach references — a character, a product, a style, a clip or a voice — and describe the scene. MuseGen blends those inputs into a single coherent clip and keeps the details you care about recognisable.
Generating needs an account and spends credits according to the model, length and resolution you pick. Each model has its own reference cap, and credits for a failed task come back on their own.
Each example starts from a character sheet, a storyboard or a visual reference we name in the brief. Read it next to the result, copy it, or push it into the generator at the top.
Reference sheet- Create a single vertical anime animation development board for an original emotional short film titled "The Girl and the Lighthouse". The output must be ONE combined image containing an anime character design sheet and a cinematic storyboard page. IMPORTANT: → Create fully original anime characters → Do not imitate any existing anime studio style directly → Do not resemble any copyrighted anime characters → Keep designs unique, soft, nostalgic, and emotionally warm STYLE: → Premium anime pre-production board → Mix loose hand-drawn storyboard sketches with semi-rendered anime keyframes → Use soft watercolor-style lighting, cool sea blue and warm lighthouse gold colors, gentle shadows, red storyboard borders, blue motion arrows, handwritten production notes, timing notes, and lens notes LAYOUT: Clean vertical composition with 2 sections. SECTION A — CHARACTER DESIGN SHEET Girl: → A gentle anime girl with long windswept dark hair, soft blue-grey eyes, and a quietly waiting expression → She wears: a soft cream knit sweater, a faded blue skirt, worn white canvas shoes, and a small green scarf → She should feel: patient, quietly hopeful, emotionally deep, nostalgically waiting → Show: front view, side profile, back view, 3/4 angle → Expressions: patient, soft smile, quietly hopeful, calm and waiting → Pose standing at the cliff edge watching the lighthouse → Pose sitting on cliff rocks with knees drawn up → Pose holding her scarf as the sea wind blows Lighthouse: → An old weathered lighthouse on a rocky seaside cliff — tall, worn stone, warm rotating golden light at the top → It should feel: steady, faithful, warm, a symbol of patient waiting and guidance → Show: wide view from cliff, close-up of rotating light, reflection on sea water Add tiny handwritten notes and soft pastel anime color swatches. SECTION B — STORYBOARD Create 8 cinematic anime storyboard panels in a neat grid. Each panel should include handwritten camera notes, blue motion arrows, timing notes, and lens notes. Story Beats: → Wide shot of an old anime seaside village at dusk — calm sea, rocky cliffs, warm lighthouse light beginning to glow → The girl walks alone along the cliff path toward the lighthouse as the sun sets → She sits down on the cliff rocks and looks out at the sea, quiet and waiting → Close-up of her face — patient, softly hopeful, emotionally still → The lighthouse light begins its slow rotation, sending warm golden light across the sea → She watches the light sweep across the water — each rotation steady and faithful → Wide shot — girl on the cliff, lighthouse above, the vast calm sea below, warm dusk light → Emotional ending — the lighthouse light sweeps across her face one more time, warm and gentle, the whole sea glowing behind her Environment: Old anime seaside village at dusk. Rocky cliffs, calm sea, old stone lighthouse. Warm golden dusk light on the water. Sea wind, distant waves, peaceful nostalgic coastal atmosphere.
Pin down the face, the outfit, the product's shape or the art direction with references, and save the prompt for the performance, camera work, pacing and sound.
Seedance 2.0 Mini is selected by default for a quick, inexpensive 480p first pass. For other reference caps, lengths and visual strengths, switch to Seedance 2.5, Gemini Omni Flash, Vidu Q3, PixVerse v6 or HappyHorse 1.1.
Four ways reference to video hands your written brief something solid to follow.
Headshots or a turnaround sheet lock in face shape, hair, clothing and proportions as the character walks into new shots and settings.
Mix images, clips and audio, then say which file governs who it is, which one the product details, and which one the camera motion, pacing or voice.
Show the real item from a few angles so its outline, materials, colours and packaging hold up inside a far bolder commercial scene.
Keep the approved subject and look while the prompt changes the performance, the location, the light and the camera path.
Assign each input a role, explain what should change, and let the model merge them.
Add the images, clips or audio that pin down your character, your product, the style, the motion or the sound. How many you can add, and of which type, depends on the model.
Write a plain brief like “Image 1 sets the face, Image 2 sets the jacket, Video 1 sets the orbiting camera,” and only then describe what happens.
Make a short take, verify the details you pinned, and run it again with a sharper rule about what stays and what moves before you commit to a longer cut.
Six workflows where staying consistent counts as much as the motion.
Keep one original character across many shots without re-describing every facial feature, outfit and colour in each prompt.
Place one photographed product in a studio set, an everyday home or a cinematic scene while its outline, surface and box stay as stable as possible.
Make quick day-in-the-life clips starring a recurring virtual person, reusing the same identity references as outfits, places and actions change.
Reference the model, the garment and the styling one by one, then direct walks, spins, close-ups and transitions into a full fashion sequence.
Let a storyboard decide framing and the order of shots, a character sheet keep continuity, and the prompt handle timing, lens moves and audio.
Reuse approved colours, materials, sets and product photos across a whole run of campaign videos instead of rebuilding the art direction each time.
Reference-driven prompts from the MuseGen gallery. Pick one, load its brief into the generator, and swap in your own files.
Reference to video runs on every model that offers it. Credits track the chosen model and length, and failed generations are refunded automatically.
For first-time AI creators
$19.9
$179 billed yearly
Save $60 compared to monthly
12,000 credits granted for the full year
Estimated monthly output
What you get
For everyday AI creation
$49.9
$419 billed yearly
Save $180 compared to monthly
30,000 credits granted for the full year
Estimated monthly output
What you get
For ambitious AI projects
$99.9
$719 billed yearly
Save $480 compared to monthly
60,000 credits granted for the full year
Estimated monthly output
What you get
Reference to video is a generation workflow that takes one or more media files together with your text brief. The references supply hard facts — who, which product, what style, how it moves, how it sounds — and the prompt says what should happen in the new clip.
Image to video usually uses a single uploaded picture as the first frame. Reference to video can give several files different creative jobs and does not force any image to be the opening frame, so it suits identity, style and multi-shot continuity better.
Seedance models can mix images, video clips and audio references. Some other models take images only or allow fewer files. The upload area changes as you switch models, so the generator only shows what that model can actually use.
Seedance 2.5 is currently set up for up to 30 images, 10 videos and 10 audio clips, capped at 50 references per project. You will rarely need that many; a handful of files with clear roles is easier to steer.
Supply clean front, profile and expression views if you can. In the prompt, say the reference governs face shape, hair, build and clothing, then list only the things that are allowed to change.
Yes. Product shots work best with several clean angles plus close-ups of key materials or packaging. Tell the model which details are fixed — above all the outline, label position, colour and surface finish.
Yes, on models that accept video references. Give the clip one clear job — body motion, shot timing or camera path — rather than asking for a full copy. That keeps the useful motion and leaves the source video's look behind.
Seedance 2.5 and Seedance 2.0 create native audio with the picture and accept audio references as well. Audio support varies by model and mode, so rely on the audio toggle shown in the generator's current settings.
Use crisp, well-lit files where the key subject is big enough to study. Skip clashing faces, strong filters and tiny product labels. Even quality across your references usually beats uploading the maximum number of files.
Your brief may not have given that file a clear role, or several references may be fighting over the same detail. Name every input directly, remove contradictions, and put the most important identity or product rule first.
Exploring the generator and setting up a project costs nothing. Rendering needs an account and credits; new accounts get 15 free credits, and failed tasks are refunded automatically.
Paid plans include commercial use under the current terms. It is still your job to hold the rights to every person, image, clip, recording, logo and product you upload as a reference.
Add the character, product, style or motion you want kept, give each file a role, and direct the new clip in a single prompt.
Build From References