LongCat Image Edit: Fast Image-to-Image Model with precise text editing on playground and API | RunComfy

What makes LongCat Image Edit stand out

LongCat Image Edit is a high-fidelity image-to-image system optimized for precise, instruction-driven changes while preserving scene structure, lighting, and identity. Built for designers, LongCat Image Edit performs targeted updates rather than full-frame resynthesis, maintaining geometry and material response for realistic outcomes. With strong text rendering and style consistency, LongCat Image Edit keeps typography legible and perspective-aware. The model balances controllability and creativity, enabling subtle retouching or global restyling without drifting from the source image.

Key capabilities:

• Structure-preserving edits: LongCat Image Edit maintains pose, layout, depth cues, and material response; avoids warping and unwanted re-synthesis.
• Consistent lighting and realism: adjusts illumination, shadows, reflections, and perspective plausibility.
• Reliable typography and signage: accurate replacement and rendering with scene-aware placement and high legibility.
• Style consistency: transfers palette and aesthetic while keeping subject identity and composition.
• Fine-grained scope control: constrain edits to background-only or object-only operations to prevent spillover.
• Deterministic iteration: seed-controlled outputs and stable behavior across batches.

Prompting guide for LongCat Image Edit

Provide a clear base image URL and a concise prompt specifying what to change and what to preserve. LongCat Image Edit responds best to concrete nouns, materials, and spatial qualifiers. Indicate global goals such as mood or style, and local scopes such as background only or keep subject untouched. Tune num_inference_steps for detail, guidance_scale for adherence, and seed for reproducibility. LongCat Image Edit supports output_format selection and acceleration settings for throughput. When iterating, LongCat Image Edit benefits from short, incremental refinements.

Examples:

• Preserve subject and composition; replace background with coastal cliffs at golden hour; keep reflections consistent.
• Only modify the sky; add dramatic storm clouds; do not alter building edges.
• In LongCat Image Edit, change car paint to deep metallic blue; preserve body lines and specular highlights.
• Use LongCat Image Edit to replace the storefront sign with "ATLAS" in white sans-serif, perspective matched.
• Add a wooden bench to the right of the subject; maintain natural shadows and scale.
• Generate 3 variations; set seed to 1234 for reproducibility; keep camera angle and framing.

Pro tips:

• Be explicit about scope and constraints: say background only or do not change the subject.
• Use spatial language: left of the subject, top-right corner, foreground only.
• Adjust control: increase num_inference_steps for fine detail; tune guidance_scale to balance adherence and creativity.
• Manage outputs: set output_format to png or webp; set acceleration to high when speed matters; leave enable_safety_checker on for compliance; use sync_mode for immediate data URIs.
• Iterate briefly: feed the latest result back into LongCat Image Edit to refine style or typography without drifting.

Note: For text to image version, use the RunComfy playground for text-to-image: LongCat Image.