Character identity
Face shape, hair, proportions and distinctive features help the subject stay recognisable across scenes. A sheet is visual context, not a guarantee of a perfect clone.
01 / Images + lip syncOne character.
One character.
More possibilities.
A character sheet gathers views, expressions or poses in one place. Use it to carry a recognisable person through a series of images, then choose one portrait to animate.
What a sheet helps you keep
Visual style
Line quality, materials, lighting and palette define the rendering style. Keep a separate style brief or reference when the character sheet is photographic but the output is illustrated.
Wardrobe and continuity
Record clothing, colours and accessories you want to repeat. Specify what may change, such as pose or setting, and what must remain fixed.
Use sheets for recurring UGC presenters, storyboards, comics, mascots and shots that need the same character. A single good portrait may be enough for one close-up. Real profile and back photographs are stronger evidence of unseen features than an invented turnaround.
Two ways to make a sheet
Composite real photographs
Put the original photographs on a plain canvas. Match their scale, keep each face large and leave space between them. This sheet adds no new visual evidence.
Download composite PNG
Generate views from one image
Upload the front portrait and ask for a consistent set of views and expressions. Inspect every panel before using it. Back views and body proportions absent from the photograph are guesses.
Character-sheet prompt
Create one clean UGC character reference sheet of the SAME adult man in the reference portrait. Ten panels in exactly five columns and two rows. Same face, brown swept hair, clean shaven, navy zip top over a white T-shirt in every panel. Top row: front portrait, left three-quarter portrait, right three-quarter portrait, left profile portrait, right profile portrait. Bottom row: neutral face, warm smile, surprised expression, upper-body back view, full-body front view. Every panel has one person only, generous space, identical warm grey background and neutral studio light. Photorealistic skin. No text, labels or borders. New angles and full body are invented reconstructions; preserve the reference identity as closely as possible.
Same brief. Four reference formats.
Does the man still look like the reference in a coffee shop? Select a brief and repetition. Compare the output first, then inspect the exact prompt and settings. All four formats use the same model, output resolution and seed for that repetition.
Change the setting
Run 1
One portrait
The front portrait alone. A useful, inexpensive baseline.
22.6s observed · $0.12 estimatedExact prompt and settings
Create a photorealistic medium portrait of this same man at a bright coffee shop table, holding a plain cream mug. Look directly at the camera. Keep his navy zip top over a white T-shirt. Natural daylight, realistic skin, one person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "2K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}Four separate images
Four real photographs, supplied as four image inputs.
31.5s observed · $0.12 estimatedExact prompt and settings
Create a photorealistic medium portrait of this same man at a bright coffee shop table, holding a plain cream mug. Look directly at the camera. Keep his navy zip top over a white T-shirt. Natural daylight, realistic skin, one person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "2K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}One composite
Those same four photographs placed on a single canvas. No new views invented.
22.7s observed · $0.12 estimatedExact prompt and settings
Create a photorealistic medium portrait of this same man at a bright coffee shop table, holding a plain cream mug. Look directly at the camera. Keep his navy zip top over a white T-shirt. Natural daylight, realistic skin, one person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "2K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}Generated character sheet
Ten views and expressions generated from one portrait. New angles are reconstructions.
22.6s observed · $0.12 estimatedExact prompt and settings
Create a photorealistic medium portrait of this same man at a bright coffee shop table, holding a plain cream mug. Look directly at the camera. Keep his navy zip top over a white T-shirt. Natural daylight, realistic skin, one person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "2K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}Judgements are temporary. Export them before reloading or leaving this page. No scores are uploaded.
Read the evidence, then choose
| Input | Outputs | Median request time | Estimated output cost |
|---|---|---|---|
| One portrait | 9 | 22.6s | $1.08 |
| Four separate images | 9 | 22.5s | $1.08 |
| One composite | 9 | 22.4s | $1.08 |
| Generated character sheet | 9 | 22.4s | $1.08 |
A polished sheet can miss the instructions
The generated sheet requested left and right profile views. Both profile panels face the same direction. Its full-body proportions and back view are also invented. Inspect every panel before treating a sheet as a reliable reference.
Style success can hide a failed brief
The cut-paper brief asked for a closed laptop. The first repetition in all four formats shows an open laptop. The style may read clearly while a concrete instruction is missed. Score those dimensions separately.
The composite contains 900 × 1200 px panels made from the 1100 px wide source derivatives. Packing and input detail change together in this practical workflow comparison; it does not isolate image count from input resolution.
Judge face similarity separately from style. A photographic result can look convincing while changing the person. A stylised result can preserve the character even when its surface appearance changes completely. Check all three repetitions before deciding.
2K output baseline
Same generated sheet, brief and seed. This tests output resolution; it does not restore detail missing from the input panels.
Exact prompt and settings
Create a photorealistic waist-up image of this same man standing in a simple warm grey studio, turned three-quarter right, pointing with his right hand at empty space on his left. Keep his navy zip top over a white T-shirt, same hair and face. One person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "2K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}4K output diagnostic
Same generated sheet, brief and seed. This tests output resolution; it does not restore detail missing from the input panels.
Exact prompt and settings
Create a photorealistic waist-up image of this same man standing in a simple warm grey studio, turned three-quarter right, pointing with his right hand at empty space on his left. Keep his navy zip top over a white T-shirt, same hair and face. One person, no writing.
{
"model": "fal-ai/nano-banana-2/edit",
"seed": 1731,
"resolution": "4K",
"aspect_ratio": "3:2",
"num_images": 1,
"output_format": "png",
"enable_web_search": false,
"limit_generations": true
}From reference sheet to talking portrait
Pick one clear, near-frontal face. Crop it out of the sheet, keep the mouth visible and supply clean speech. The sheet is a reference for making a portrait, not a canvas of ten people to animate together.
Original portrait · neutral
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"references/front.png",
"audio/voice-neutral-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.999
}
}Automated video review
Lip sync follows the cadence with natural mouth shapes and steady eye contact. Head movement is subtle.
Minor edge warping along the jawline
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Original portrait · excited
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"references/front.png",
"audio/voice-excited-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.937
}
}Automated video review
Mouth opens wider matching emphatic delivery; eye contact remains steady with slightly wider gaze.
Slight silhouette flickering on left collar
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Original portrait · reassuring
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"references/front.png",
"audio/voice-reassuring-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.942
}
}Automated video review
Calm articulation with smooth closures on bilabials. Subtle blinks and consistent facial geometry.
Subtle boundary artifact near neckline
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Generated portrait · neutral
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"images/setting-generated-1.png",
"audio/voice-neutral-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 5.037
}
}Automated video review
Mouth timing corresponds to speech, though torso and head remain largely static. Teeth appearance is slightly softened.
Soft focus around mouth interior; Unnatural upper body stillness
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Generated portrait · excited
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"images/setting-generated-1.png",
"audio/voice-excited-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 5.038
}
}Automated video review
Mouth opens broader for the energetic audio, but upper face and eyes do not fully reflect the excitement.
Minor teeth smearing during open vowels
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Generated portrait · reassuring
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"images/setting-generated-1.png",
"audio/voice-reassuring-8s.mp3"
],
"resolution": "768P",
"enable_transcription": true,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.97
}
}Automated video review
Gentle lip shapes align well with audio pacing. Identity stays consistent with source cafe image.
Slight temporal blur on lower lip
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Original portrait · neutral · transcript guidance off
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"references/front.png",
"audio/voice-neutral-8s.mp3"
],
"resolution": "768P",
"enable_transcription": false,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.959
}
}Automated video review
Reliable phoneme tracking and realistic jaw displacement. Blinking and gaze hold naturally.
Faint matte shimmer on neck edge
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Generated portrait · neutral · transcript guidance off
Compare mouth timing, teeth, facial identity and movement. The same audio is used for both portrait sources in this delivery.
Download MP4Transcript and provenance
Before you press publish, pause. Check the face, check the voice, and make sure the character still feels like the same person.
The clip uses up to the first eight seconds of this script; the audio waveform determines its actual ending.
{
"model": "minimax/h3-max/lip-sync/image-to-video",
"inputs": [
"images/setting-generated-1.png",
"audio/voice-neutral-8s.mp3"
],
"resolution": "768P",
"enable_transcription": false,
"seed": 1731,
"enable_safety_checker": true,
"timings": {
"inference": 4.951
}
}Automated video review
Accurate syllable timing with preserved facial identity, accompanied by fixed posture.
Subtle smoothing around oral cavity
Automated review of a 512 px video derivative. This does not certify exact lip timing or a human review.Two extra neutral clips switch transcript guidance off while keeping the portrait, audio and seed fixed. Compare them with their guidance-on counterparts before deciding whether this API setting helps your input.
Your turn: choose a reference strategy
- Download the reference pack, or use photographs you have permission to use.
- Make a composite without inventing new angles. Generate a second sheet from one portrait.
- Keep one brief and model fixed. Generate at least three repetitions for each input strategy.
- Score likeness, wardrobe, style and brief adherence separately. Include failed attempts.
- Choose a strategy for a three-image campaign and explain the quality, preparation effort and cost tradeoff.
Success: you can explain why you chose a reference format using a repeatable comparison. A beautiful single result is not enough evidence of consistency.
Next: assemble a local animation →