Custom Image Generation v2

Image·eachlabs·by eachlabs

By combining an input image with preset styles, Custom Image Generation v2 generates refined images that retain the original layout with stylistic updates.

Try it now →

API reference

Runtime (p50): 3m
Estimated price: $0.08

Call the API

prediction.sh

curl -X POST \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "custom-image-generation-v2",
    "version": "0.0.1",
    "input": {
        "image_url_1": "https://storage.googleapis.com/magicpoint/inputs/custom-image-generation-v2.png",
        "style_slug": "ghibli-anime"
    },
    "webhook_url": ""
}' \
  https://api.eachlabs.ai/v1/prediction/

Documentation8 sections

Overview
Custom Image Generation v2 — Image-to-Image AI Model

Custom Image Generation v2 from Eachlabs revolutionizes image-to-image AI model workflows by blending an input image with preset styles to produce refined outputs that preserve the original layout while applying precise stylistic updates. Developed within the Eachlabs family, this model excels in maintaining structural integrity during transformations, making it ideal for developers seeking an image-to-image AI model that delivers consistent, high-fidelity edits without distorting core compositions. Whether enhancing product photos or redesigning creative visuals, Custom Image Generation v2 ensures outputs retain essential geometry and details, streamlining tasks like automated image editing for e-commerce and professional design iterations.
Capabilities
- Transforms input images by applying a wide range of preset styles while retaining core composition and layout
- Supports iterative refinement, enabling users to progressively improve outputs through parameter adjustments
- Produces high-quality, visually coherent images suitable for professional and creative applications
- Offers flexible control over style strength, prompt guidance, and randomness for tailored results
- Capable of upscaling, retouching, and outpainting for extended creative workflows
- Maintains consistency across multiple generations when using the same seed and settings
Use cases
Use Cases for Custom Image Generation v2

For designers prototyping UI mockups, upload a wireframe image and apply a "modern glassmorphism style with neon accents" preset to generate polished variants that keep layout fidelity, accelerating iteration without redrawing elements—ideal for rapid edit images with AI workflows.

Marketers handling e-commerce visuals can input product shots with a prompt like "apply luxurious gold-toned studio lighting while retaining shelf arrangement and shadows," producing catalog-ready images that boost appeal without reshooting, leveraging the model's strength in AI image editor API for scalable edits.

Developers building apps integrate Custom Image Generation v2 via API to enable user-driven style swaps on uploaded photos, such as transforming casual portraits into "professional headshots with corporate backdrop," ensuring identity consistency for avatar generators or social tools.

Content creators refining assets use multi-reference inputs to blend character designs across poses with a unified anime style preset, maintaining proportions for comic series or game art, streamlining production in automated image editing API pipelines.
Tips & tricks
How to Use Custom Image Generation v2 on Eachlabs

Access Custom Image Generation v2 seamlessly on Eachlabs via the Playground for instant testing, API for production-scale Custom Image Generation v2 API integrations, or SDK for custom apps. Provide an input image, optional text prompt up to 10,000 characters, style presets, and reference images; select resolution up to 4MP and steps for quality-speed balance. Outputs deliver high-fidelity PNG/JPG images with preserved layouts in seconds.
---
Technical spec
What Sets Custom Image Generation v2 Apart

Custom Image Generation v2 stands out in the competitive landscape of eachlabs image-to-image tools through its specialized focus on layout-preserving style transfers, distinguishing it from general-purpose editors that often alter compositions. It supports high-resolution outputs up to 4 megapixels, with input formats including JPG, PNG, and multiple reference images for context-aware refinements. Average processing time is sub-second on optimized hardware like RTX 5090, balancing speed with quality via configurable steps from 1-50.
- Layout retention with stylistic overlays: Combines input images with preset styles to update aesthetics without changing poses or structures; this enables seamless batch editing for e-commerce catalogs where product positioning must remain identical across variants.
- Multi-reference image support: Incorporates elements from several input images into coherent outputs; users gain consistent character or branding across edited series, perfect for AI photo editing for e-commerce.
- Efficient high-res editing: Handles up to 4MP resolutions with preserved details and low VRAM needs (8-13GB); this allows professional-grade transformations on consumer GPUs, unlike resource-heavy alternatives.
Things to be aware of
- Some users report that extreme style influence can distort key features or introduce unwanted artifacts
- Performance and output quality are highly dependent on hardware; slower generation on lower-end GPUs
- Consistency across batches is generally strong when using fixed seeds, but minor variations can occur
- Community feedback highlights the ease of iterative refinement and the usefulness of side-by-side comparisons
- Positive reviews often mention the model’s ability to preserve layout and recognizable elements even with strong stylistic changes
- Negative feedback sometimes notes occasional mismatches between prompt intent and output, especially with ambiguous or conflicting instructions
- Resource requirements can be significant for high-resolution outputs or large batch generations
Key considerations
- Adjust the style influence slider to control how much the preset style affects the final image versus the original input
- Higher sampling steps generally yield more detailed and higher-quality images but increase generation time
- Guidance scale determines how closely the output adheres to the prompt and style; fine-tune for best results
- Use seeds to ensure reproducibility or introduce controlled randomness for variation
- Iterative refinement (generating, inspecting, and tweaking parameters) is essential for achieving production-ready visuals
- Balance between creativity and consistency by experimenting with prompt structure and style strength
- Avoid overly complex or conflicting prompts, as these can reduce output quality or introduce artifacts
Limitations
- May struggle with highly complex or abstract prompts that conflict with the original image’s structure
- Output quality and speed are limited by available hardware, especially for high-resolution or batch processing
- Not optimal for generating entirely new compositions from scratch; best suited for style transfer and refinement of existing images