Soup.video

INPUT_IMAGE: use the provided image as the only source; do... by @ayush_bhardwaj - soup.video

INPUT_IMAGE: use the provided image as the only source; do not replace or alter composition, objects, or facial identity. TASK: restore, repair, colorize if...

INPUT_IMAGE: use the provided image as the only source; do not replace or alter composition, objects, or facial identity. TASK: restore, repair, colorize if needed, denoise, and upscale the image to 3840x2160 (4K) while preserving every original element, pose, expression, and color identity. The model must repair cracks, scratches, and compression damage; reconstruct missing micro-details plausibly; enhance perceptual detail and texture; and output a natural, realistic result without adding new objects or changing identity. ANALYSIS PHASE: - Detect image condition: identify cracks, scratches, tears, heavy noise, compression blocks, blur, and whether image is grayscale or color. - Determine appropriate restoration pipeline steps and confidence levels for each region. RESTORATION PHASE: - Repair physical damage: inpaint cracks and scratches using context-aware reconstruction that matches surrounding texture and grain. - Denoise and remove JPEG/blocking artifacts while preserving original grain and texture characteristics. - Reconstruct lost micro-details (skin pores, hair strands, fabric weave, surface reflections) using detail-aware synthesis constrained by surrounding pixels. - If image is grayscale/black-and-white: perform **context-aware colorization** that infers natural skin tones, fabric colors, and environmental colors; preserve plausible historical/scene accuracy and avoid invented logos or text colors. - If image is already color: **do not change original color values** except to remove color noise; preserve hue and overall color identity. - Local contrast and clarity: selectively increase midtone contrast and local clarity on subject and important details without clipping highlights or crushing shadows. - Edge and texture sharpening: apply perceptual, artifact-aware sharpening to fine edges; avoid halos and plastic look. - Global consistency: maintain original lighting direction and highlight/shadow relationships; do not relight the scene. UPSCALE PHASE: - Detail-aware upscaling to 3840x2160 (4K) with artifact-free reconstruction; synthesize plausible high-frequency detail only where supported by context. - Preserve facial feature fidelity and proportions; avoid smoothing or altering identity. Maintain original composition and do not add or remove objects, text, logos, or background elements. OUTPUT REQUIREMENTS: - Output: high-quality 4K image (3840x2160) in PNG or high-quality JPEG. - Provide optional: a short changelog listing applied local operations (areas inpainted %, colorized yes/no, denoise strength, upscaling method). - If colorization performed: include a brief color rationale file describing choices for skin, clothing, and background. NEGATIVE INSTRUCTIONS: - Do not change pose, expression, facial features, or proportions. - Do not add, remove, or replace objects, logos, or background elements. - Do not invent or paste recognizable elements (faces, text, objects) not present in the source. - Do not apply stylized painting, cartoon, or illustrative effects. - Do not oversaturate, shift hue globally, or apply aggressive color grading. - Do not produce visible sharpening halos or plastic skin. RECOMMENDED PARAMETERS: - Resolution: 3840x2160 - Denoise strength: 0.20–0.45 (adjust per-image) - Inpaint/repair aggressiveness: adaptive; start low and increase only where cracks exist - Upscale method: detail-aware reconstruction / context-preserving - Sharpening: moderate, artifact-aware - Steps: 25–45 - Guidance/CFG scale: 6–12 - Face restoration: enabled only if it preserves identity - Output format: PNG 24-bit or high-quality JPEG
@ayush_bhardwajAuthor
Prompt
INPUT_IMAGE: use the provided image as the only source; do not replace or alter composition, objects, or facial identity. TASK: restore, repair, colorize if needed, denoise, and upscale the image to 3840x2160 (4K) while preserving every original element, pose, expression, and color identity. The model must repair cracks, scratches, and compression damage; reconstruct missing micro-details plausibly; enhance perceptual detail and texture; and output a natural, realistic result without adding new objects or changing identity. ANALYSIS PHASE: - Detect image condition: identify cracks, scratches, tears, heavy noise, compression blocks, blur, and whether image is grayscale or color. - Determine appropriate restoration pipeline steps and confidence levels for each region. RESTORATION PHASE: - Repair physical damage: inpaint cracks and scratches using context-aware reconstruction that matches surrounding texture and grain. - Denoise and remove JPEG/blocking artifacts while preserving original grain and texture characteristics. - Reconstruct lost micro-details (skin pores, hair strands, fabric weave, surface reflections) using detail-aware synthesis constrained by surrounding pixels. - If image is grayscale/black-and-white: perform **context-aware colorization** that infers natural skin tones, fabric colors, and environmental colors; preserve plausible historical/scene accuracy and avoid invented logos or text colors. - If image is already color: **do not change original color values** except to remove color noise; preserve hue and overall color identity. - Local contrast and clarity: selectively increase midtone contrast and local clarity on subject and important details without clipping highlights or crushing shadows. - Edge and texture sharpening: apply perceptual, artifact-aware sharpening to fine edges; avoid halos and plastic look. - Global consistency: maintain original lighting direction and highlight/shadow relationships; do not relight the scene. UPSCALE PHASE: - Detail-aware upscaling to 3840x2160 (4K) with artifact-free reconstruction; synthesize plausible high-frequency detail only where supported by context. - Preserve facial feature fidelity and proportions; avoid smoothing or altering identity. Maintain original composition and do not add or remove objects, text, logos, or background elements. OUTPUT REQUIREMENTS: - Output: high-quality 4K image (3840x2160) in PNG or high-quality JPEG. - Provide optional: a short changelog listing applied local operations (areas inpainted %, colorized yes/no, denoise strength, upscaling method). - If colorization performed: include a brief color rationale file describing choices for skin, clothing, and background. NEGATIVE INSTRUCTIONS: - Do not change pose, expression, facial features, or proportions. - Do not add, remove, or replace objects, logos, or background elements. - Do not invent or paste recognizable elements (faces, text, objects) not present in the source. - Do not apply stylized painting, cartoon, or illustrative effects. - Do not oversaturate, shift hue globally, or apply aggressive color grading. - Do not produce visible sharpening halos or plastic skin. RECOMMENDED PARAMETERS: - Resolution: 3840x2160 - Denoise strength: 0.20–0.45 (adjust per-image) - Inpaint/repair aggressiveness: adaptive; start low and increase only where cracks exist - Upscale method: detail-aware reconstruction / context-preserving - Sharpening: moderate, artifact-aware - Steps: 25–45 - Guidance/CFG scale: 6–12 - Face restoration: enabled only if it preserves identity - Output format: PNG 24-bit or high-quality JPEG
Model
Qwen Image 2.0
Type
Image
Ratio
16:9
Visibility
public
Created
May 15, 2026