Docs Menu

ImageGPT Model

A practical guide with ready-made prompt templates, examples, and tips.

Upload up to three photos, formulate short commands (Add / Remove / Replace / Change / Transform / Move), and always specify what to change, where, and how much.


What the Model Can Do

  • Multi-editing (1-3 input images): combine "person + scene", "person + object/product", "person + person", etc.
  • Enhanced consistency: more careful with faces, branding, and text (yes, your corporate font will stop suffering from "creative freedom").
  • Targeted edits: add/remove/replace objects, change background/pose/material, replace text while preserving style.
  • Context understanding: correctly interprets short text instructions and maintains character identity even through a series of edits.

ℹ️ Results are more stable with English prompts. Russian translations are provided below for convenience, but final prompts are better written in English.

⚠️ Practical tip: plan one key edit per call. For a series of edits - perform a step → evaluate the result → submit as a new input image. Patience is a superpower.


Quick Prompt Formulation Rules

  1. "What not to change" first, then - the edit. Example: "Keep face, hairstyle, proportions, lighting and background. Replace ..." - this drastically reduces drift. The model loves clarity. Like cats love boxes.

  2. Command verbs are your friends: Add / Remove / Replace / Change / Move / Transform. Short phrases are better than long poems. Poetry is great, but not here.

  3. Text in frame - in double quotes and with an explicit command: Replace "OLD" to "NEW"; if necessary, specify location, size, color, font material.

  4. For 3 images call them Image 1 / Image 2 / Image 3 and clearly state what to take from where: "Make the woman from Image 1 wear the dress from Image 2 and the pose from Image 3".

  5. Location and quantity - as specific as possible: "in the bottom-right corner", "one item", "on the left shoulder", "size ~20% of frame".

  6. Negative prompt keep short (or empty) - large lists of prohibitions often interfere.


  1. Upload 1-3 images. We recommend the pattern: Image 1 - base (where we make changes), Image 2/3 - donors (objects, style, pose).

  2. Compose a prompt using the formula:

  • "Keep" section: what's important not to touch.
  • "Do/Replace" section: short commands + object/attribute + position/quantity.
  • "Sources" section: if taking something from additional images - specify Image 2/3.
  1. Run generation and evaluate the result. For a chain of edits - repeat the process with the fresh result.

Ready-Made Templates

Targeted Edits (one photo)

  • Color/material change
Keep face, pose and lighting.
Change the jacket to dark-green leather with a subtle sheen.
Keep everything else unchanged.
  • Object removal
Keep the composition and background.
Remove the blue graffiti text on the wall.
Do not alter bricks, lighting or perspective.
  • Background replacement
Keep face, hairstyle and outfit.
Transform background to an evening city skyline with bokeh.
Camera remains frontal, same framing.

Working with Text in Frame

  • Text replacement
Replace "HEALTH INSURANCE" to "Tomorrow will be better",
keep typography coherent with current style, place centered on the blocks.
  • Add with parameters
Add text "LIMITED EDITION" at the top-center,
slight drop shadow, match existing font weight.

Multi-editing (2-3 photos simultaneously)

  • Dress + pose
Image 1 is the base.
Make the woman from Image 1 wear the black dress from Image 2
and sit in the pose from Image 3.
Preserve face, hair, lighting and camera angle.
  • Product + poster/scene
Image 1 is the product photo, Image 2 is the target scene.
Place the product from Image 1 on the table in Image 2,
cinematic studio lighting, realistic reflections.
Keep proportions and brand details unchanged.
  • Face swap (reference portrait + scene)
Replace the person's face in Image 1 with the face from Image 2.
Keep hairstyle, body pose, clothing and background unchanged.
Blend skin tone and lighting naturally.

Style/Genre

  • Photo → anime (preserving appearance)
Keep facial features, hairstyle and outfit.
Transform to a high-quality modern anime style with smooth digital shading and glowing highlights.
Keep the background and framing the same.
  • Post-processing/restoration
Restore old photograph, remove scratches, reduce noise, enhance details,
high resolution, realistic, natural skin tones, clear facial features, no distortion.

Restoration and Colorization

  • Full restoration + 4K quality
Restore and colorise this picture. Remove any imperfections and make it look like a 4k photo.
  • Preserve scene, remove damage
Restore this photo to a fresh state, preserving the original scene but removing any damage or degradation.
  • Remove artifacts and tears
Remove scratches, dust spots, noise and fill in any ripped sections, turning it into a high quality photograph.
  • Colorize B&W + enhance quality
Colorize this black and white photograph and enhance the overall quality.

Artistic Effects

  • Cyberpunk with neon
Transform to cyberpunk style with neon lighting.
  • Pencil sketch
Convert to pencil sketch with cross-hatching and visible paper texture.
  • Pixar-like 3D animation
Make this look like a Pixar 3D animation.
  • Claymation (clay)
Transform to clay sculpture style (claymation).

Backgrounds and Scenes

  • Beach, keep subject
Change the background to a sunny beach while keeping the person in exact same position, scale, and pose.
  • Night cyberpunk city
Replace background with cyberpunk city at night, maintain identical subject placement.
  • Remove background figure
Remove the person in the background while keeping everything else unchanged.

Anti-patterns and How to Fix

  • Too vague: "make it beautiful, retro style". How to do it: specify style through features: "1970s: disco ball, mirrored walls, saturated colors, warm light".

  • No position/quantity: "add a cat". How to do it: "Add a light-gray cat in the bottom-right corner, sitting, facing the camera, one cat."

  • Overloaded negative prompt: interferes with the task. How to do it: short negative (or empty).


Mini-FAQ

How many photos can I upload? Up to three. Optimal - 1-3 input images. No need for more (and the service will find it hard to carry so many suitcases).

How to correctly reference uploaded photos in the prompt? Call them Image 1 / Image 2 / Image 3 and formulate: "Take [object] from Image 2, place on Image 1 ...".

Can I make several sequential edits in one call? We recommend one key change at a time - more stable and predictable result. For a series of edits, simply submit the result back as a new input image.


Prompt Skeleton Template

[SAVE] Keep {list what to preserve: face, hairstyle, proportions, lighting, background, framing}.

[DO] {Add/Remove/Replace/Change/Transform/Move} {what exactly}
{attributes: color/material/size/pose/style}
{position/quantity, if important}.

[SOURCES] If using multiple images:
Use {object/style/pose} from Image 2 {and/or} Image 3.
Keep everything else unchanged.

Example (3 photos):

Keep face, hair, outfit, framing and lighting.
Replace the dress on the woman with the black dress from Image 2
and set the pose from Image 3 (seated, profile 3/4).
Keep everything else unchanged.

Substitute your objects/styles and go. If the prompt reads like a good instruction for a designer assistant - you're on the right track.

Useful "What to Keep" Phrases

Use short English templates to lock down what's important:

while maintaining facial features
preserve the original character appearance
keep everything else unchanged
maintain identical composition

Note on Rights and Ethics

Please respect copyright, brand guidelines, and privacy of people in photos. Don't create misleading images and don't imitate people without their consent. Let's be cool - and the world will be a little better.


Done! If you want, we can package this guide into your brand guide, add hints directly to the interface (placeholders in the input field and help tooltips at the 3-image uploader) - and it will be perfect.


Professional Use Cases (e-commerce, architecture, portrait)

  • Studio set for product
Place this product in elegant studio setting with professional lighting.
  • Archviz: material and greenery
Change building materials from concrete to red brick, add ivy growing on walls, maintain architectural proportions.
  • LinkedIn portrait
Transform into professional LinkedIn headshot with business attire and clean background, maintain exact facial features.

Limitations and Tips from Experienced Users

Model Limitations

  • After a series of >6 sequential edits, artifacts are possible - break work into steps and more often "save as a new base image".
  • Specific instructions are sometimes ignored - help with clarifying attributes: size, position, quantity.
  • English prompts work more stably - use English for final formulation.

Workflow Tips

  • Do one key change at a time, keep the original for a series of experiments.
  • To save credits, reduce batch size to 1-2.

Advanced Techniques

  • For text replacement use quotes: Replace "SALE" with "SOLD".
  • Perform complex transformations step by step (each step - separate call).
  • For stylization try reference images and style descriptions, but lock down what to keep (face, pose, lighting, composition).