Prompting Guide - Image-to-Image

Overview This guide consolidates best practices for editing images using Kontext-style image-to-image workflows.

​

Basic Object Modifications

Kontext is effective for straightforward object modifications such as recolors, replacing objects, or minor retouching. Example prompt: “Change the color of the yellow car to deep cherry red while preserving reflections and highlights.” [Placeholder image: Input image — add image here] [Placeholder image: Output image — add image here]

​

Prompt Precision: From Basic to Comprehensive

​

Quick Edits

Simple prompts can work but may alter style or composition. Prompt example: “Change to daytime” [Placeholder image: Input image for quick edit] [Placeholder image: Output 1] [Placeholder image: Output 2]

​

Controlled Edits

Add preservation instructions to keep style and composition similar to the input. Prompt example: “Change to daytime while maintaining the same style of the painting” [Placeholder image: Input image for controlled edit] [Placeholder image: Controlled edit output]

​

Complex Transformations

For multiple simultaneous changes include clear, ordered instructions and prioritize the most important changes. Prompt example: “Change the setting to daytime, add several people walking on the sidewalk, keep the original painting style and composition” [Placeholder image: Input image for complex transform] [Placeholder image: Complex transform output]

​

Style Transfer

Use direct style names, artist references, and descriptive characteristics.

​

Using textual style prompts

1. Name the specific style (e.g., “Bauhaus”, “watercolor”, “film noir”)
2. Reference artists or movements when appropriate
3. Describe key characteristics: brushstrokes, color palette, texture
4. Preserve composition if needed: “keep original composition and object placement”

[Placeholder image: Architectural photo input]

​

Using an input image as a style reference

Provide the style image as a reference and then describe the content you want in that style. Prompt example: “Using this style reference image, create a scene where a bunny, a dog, and a cat are having a tea party around a small white table.” [Placeholder image: Style reference] [Placeholder image: Generated output using style reference]

​

Iterative editing & character consistency

Kontext preserves character identity well when prompts explicitly request preservation. Framework to maintain character consistency:

• Establish the reference: “The woman with short black hair and a mole on her left cheek…”
• Specify the transformation: environment, activity, or style
• Preserve identity markers: “maintain the same facial features, hairstyle and expression”

Example prompts in editing sequences:

1. “Remove the sunglasses from the woman’s face while keeping expression unchanged”
2. “Place the same woman in a snowy street while preserving facial features and pose”

[Placeholder image: Reference character] [Placeholder image: Iteration step 1 output] [Placeholder image: Iteration step 2 output]

​

Text Editing in images

Use quotation marks around exact text you want to change. Prompt structure: Replace ‘[original text]’ with ‘[new text]’ Example: “Replace ‘Choose joy’ with ‘Choose BFL’ while maintaining original font, color and size” [Placeholder image: Sign with text ‘Choose joy’] [Placeholder image: Sign changed to ‘Choose BFL’] Best practices for text edits:

• Use exact punctuation and casing
• Ask to preserve font, color, size, and layout when necessary
• Keep replacement text similar in length to avoid layout issues

​

Visual cues & masks

Use visual markers, masks, or bounding descriptions to indicate where edits should occur. Example: “Add hats inside the three boxes drawn on the upper right quadrant” [Placeholder image: Input with boxes] [Placeholder image: Output with hats added]

​

Troubleshooting: When results don’t match expectations

• If the model changes parts you wanted preserved, explicitly state what should remain unchanged: “Keep everything else in the image identical”
• For character identity drift, enforce identity markers: “preserve exact facial features, hairstyle, eye color”
• If composition shifts unintentionally, state: “Keep the subject in the exact same position, scale, and pose”

​

Composition control

Vague prompts like “put him on a beach” can change framing and camera angle. Prefer: “Change the background to a sunny beach while keeping the person in the exact same position, scale, pose, camera angle, framing and perspective. Only replace the environment around them.” [Placeholder image: Composition input] [Placeholder image: Composition-preserved output]

​

Style not applying correctly

Use richer style descriptions: “Convert to pencil sketch with natural graphite lines, cross-hatching, and subtle paper texture” [Placeholder image: Input photo] [Placeholder image: Precise sketch output]

​

Safety & Content Guidelines

• Avoid requesting generation of disallowed content (follow your platform’s content policy)
• Obfuscate or avoid personal identifying edits if you do not have consent

​

Best Practices Summary

• Be specific: use exact descriptors for color, lighting, and materials
• Start simple: make incremental changes and iterate
• Preserve intentionally: call out what must not change
• Use quotes for text edits: “Replace ‘X’ with ‘Y’”
• Control composition explicitly: specify camera angle, framing, and subject placement
• Choose verbs carefully: “transform” often implies full replacement; “change the clothes” is more focused

[Placeholder section for example prompts and executions — add example images and model responses here]