Brand images with AI

The demo

The system

Base reference

Higgsfield / GPT Image 2one image on white, the official face

↓one image rules: everything else comes from it

Character sheet

🧑‍🎨Multi-angle sheet6 views of the same face

↓this is the consistency lock

Scenes

🧥Outfitsthe brand's wardrobe

🎬Locationseach world, same person

↓the character travels, the identity doesn't change

Consistency control

🎯Human eyecompares against the base, rejects the drift

The tools

Higgsfieldgeneration

The engine where I generate characters and scenes with director prompts. It keeps the face across images.

Try it ↗

ChatGPT (GPT Image 2)base image

The other route for the base reference: the casting photo you attach is the character's official face.

Try it ↗

📌

Pinterestreference

Where I pull the art direction from: light, framing and mood before writing the prompt.

Learn more ↗

⚡

APImartgeneration

The same models, via API and at a better cost. My route to scale volume.

Try it ↗referral

Replicate it step by step

The code, the connections and the mistakes I already made for you.

1. 1

   Write the character brief before generating anything

   Consistency starts in a document, not in a prompt. Before touching the tool, I define the character's fixed traits: face, skin tone, hair, distinctive marks and the base wardrobe. If you can't describe the person in one paragraph, the model will invent a different one every time. This is the brief I fill:

   checklist · character briefCopy

   CHARACTER BRIEF (the traits that NEVER change):
   - Name and role within the brand.
   - Face: shape, apparent age, identifying features.
   - Skin and hair: tone, texture, fixed cut or style.
   - Distinctive marks: what makes them recognizable at a glance.
   - Base wardrobe: the neutral outfit of the reference.
   - Brand rule: plain garments or your brand only, never third parties.
   This brief gets pasted inside the base reference prompt.

2. 2

   Generate the base reference on white

   The base reference is a single image on a white background, full body, even light and neutral clothing. It's the official face: every future image gets compared against this one. Prompts are in English on purpose: image models understand them better and the results come out more consistent. Here I paste the character brief and turn on the photoreal stack so it looks like a photo, not a render:

   prompt · base referenceCopy

   Create a single full-body studio reference of one fictional character on a
   clean white seamless background. This image is the master reference: every
   future image of this character must match this exact face and identity.
   
   Character: [paste your character brief here, step 1].
   Framing: full body, centered, neutral standing pose, arms relaxed, eyes into
   the lens. Even, soft studio light. Plain wardrobe in one solid color, no logos,
   no third-party brands.
   
   Photoreal stack (keep all of it): visible skin pores and fine texture,
   subsurface scattering, strand-by-strand hair, natural fabric weave, shot on
   Kodak Vision3 color, 50mm lens, honest depth of field. No beauty filter,
   no plastic skin, no cinematic color grading.
   
   Negative prompt: cartoon, illustration, anime, plastic skin, airbrushed,
   extra fingers, deformed hands, warped face, text, logos, third-party brands,
   watermark, low quality.

3. 3

   Pull the multi-angle character sheet

   With the base approved, I generate a 6-view sheet of the same person: front, three-quarter, profile, back and a close-up of the face. This sheet is the consistency anchor: it's what I attach as a lock every time I take the character into a new scene.

   prompt · character sheetCopy

   Use the attached base reference as the only source of truth for identity.
   Keep the face, skin tone, hair and proportions 100% identical to it.
   
   Build a 6-panel character sheet of the SAME person on the same white seamless
   background: front, three-quarter left, profile, three-quarter right, back, and
   one chest-up close of the face. Same wardrobe, same light across all six panels.
   This sheet is the consistency anchor for every scene later.
   
   Keep the full photoreal stack from the reference (pores, subsurface, strand
   hair, Kodak Vision3, no beauty filter). No text labels baked into the image,
   no logos, no third-party brands.

4. 4

   Take the character into scenes

   Now the character travels. I attach the sheet as the identity lock and only change the world: an outfit from the wardrobe, a location, light and mood. The one-brand rule goes explicit in the prompt, because otherwise the model adds third-party logos on its own.

   prompt · scene with the characterCopy

   Use the attached character sheet as the identity lock. The person must stay
   exactly the same: face, skin, hair, body. Do not invent a new person.
   
   Place the SAME character into this scene: [describe the setting, the outfit
   from your wardrobe, the light and the mood]. Cinematic but honest framing,
   one clear subject, negative space for text if this is a cover.
   
   Wardrobe rule: plain garments or your own brand only, never third-party logos.
   Keep the photoreal stack (pores, subsurface, strand hair, natural fabric,
   Kodak Vision3, no beauty filter, no plastic skin).
   
   Negative prompt: different face, identity drift, plastic skin, extra fingers,
   deformed hands, third-party brand logos, text artifacts, watermark, low quality.

5. 5

   Consistency control: the human gate

   The last filter is human and it's what holds everything together: each scene gets compared against the base reference and the character sheet. If the face drifted, it gets rejected and regenerated. Consistency doesn't come from the text prompt, it comes from always anchoring to the same visual reference. byTHELAB keeps its own cast of recurring characters precisely to prove this is directable, not a matter of luck.

Keep going with these