THIS ARTICLE IS OUTDATED, PLEASE USE THIS ONE:

BEST NEGATIVE PROMPT FOR NOOBAI-XL

Main tags

These tags have been used by developers and have been tested by me personally they work and improve the generation results.

A complete list of the main tags:

worst quality, low quality, bad quality, lowres, low res, pixelated, blurry, blurred, compression artifacts, jpeg artifacts, downscaled, bad anatomy, bad hands, worst hands, deformed hands, deformed fingers, deformed feet, deformed toes, extra limbs, extra arms, extra legs, extra legs and arms, extra fingers, extra digits, extra digit, mutated hands, mutation, mutated, fused fingers, fused toes, missing limbs, missing arms, missing fingers, missing toes, malformed, disfigured, wrong hands, ugly hands, ugly fingers, twisted hands, flexible deformity, conjoined, disembodied, text, watermark, signature, logo, url, username, tagme, bad, ugly, worst, very displeasing, displeasing, error, doesnotexist, unfinished, poorly drawn face, poorly drawn hands, poorly drawn feet, artistic error, bad proportions, bad perspective, bad foreshortening, bad twitter id, censored, bar censor, mosaic censoring, out of frame, cropped, missing, missing id, bad id, ai-generated, ai-assisted, stable diffusion, overly saturated, overly vivid, cross-eye, expressionless, extra, fewer, early, old, oldest, scan, sketch, manga, comic, monochrome, simple background, abstract, sequence, lineup, turnaround, 2koma, 4koma, what, microsoft paint \(medium\), artifacts, adversarial noise, has bad revision, scan artifacts, stitched, resized, image sample,

• Quality-Related Negatives: These tags are crucial as they directly affect the overall image quality. Removing low-resolution, blurry, or pixelated outputs ensures the generated images are sharp and detailed.

worst quality, low quality, bad quality, lowres, low res, pixelated, blurry, blurred, compression artifacts, jpeg artifacts, downscaled

• Anatomy-Related Negatives: AI models often struggle with human anatomy, especially hands and limbs. Including these tags helps avoid distorted or unnatural body parts, improving the realism of the image.

bad anatomy, bad hands, worst hands, deformed hands, deformed fingers, deformed feet, deformed toes, extra limbs, extra arms, extra legs, extra legs and arms, extra fingers, extra digits, extra digit, mutated hands, mutation, mutated, fused fingers, fused toes, missing limbs, missing arms, missing fingers, missing toes, malformed, disfigured, wrong hands, ugly hands, ugly fingers, twisted hands, flexible deformity, conjoined, disembodied

• Unwanted Text and Artifacts: Text, watermarks, and signatures can distract from the main subject. Negating these elements keeps the focus on the desired content.

text, watermark, signature, logo, url, username, tagme

• General Negative Descriptors: These tags filter out images with common flaws or unattractive features, enhancing the aesthetic quality.

bad, ugly, worst, very displeasing, displeasing, error, doesnotexist, unfinished, poorly drawn face, poorly drawn hands, poorly drawn feet, artistic error, bad proportions, bad perspective, bad foreshortening, bad twitter id

• Censorship Artifacts: Censor bars and mosaics can obscure important details. Excluding these ensures the image is clear and unobstructed.

censored, bar censor, mosaic censoring

• Framing and Composition Errors: Proper framing is essential. These tags prevent issues like subjects being cut off or important elements missing from the frame.

out of frame, cropped, missing, missing id, bad id

• AI-Related Negatives: Removes elements specific to AI-generated images, such as artifacts or styles tied to certain models, leading to more natural results.

ai-generated, ai-assisted, stable diffusion

• Color and Saturation Issues: Overly saturated or vivid colors can look unnatural. These tags help in achieving a balanced color palette.

overly saturated, overly vivid

• Miscellaneous Errors: These are less critical but still impact the overall quality. They include style preferences (e.g., avoiding comic or sketch styles) and other minor flaws.

cross-eye, expressionless, extra, fewer, early, old, oldest, scan, sketch, manga, comic, monochrome, simple background, abstract, sequence, lineup, turnaround, 2koma, 4koma, what, microsoft paint \(medium\), artifacts, adversarial noise, has bad revision, scan artifacts, stitched, resized, image sample,

Extra tags

These tags have not been tested and were written by the OpenAI o1-preview model, most likely some of them work (like styles) but I wouldn't use them without further testing.

A complete list of the extra tags:

asymmetrical eyes, distorted face, deformed face, incorrect facial features, bad face, extra eyes, extra mouth, missing eyes, missing mouth, mismatched facial features, unnatural pose, awkward pose, dislocated limbs, stretched limbs, shrunken limbs, floating limbs, disconnected limbs, distorted limbs, asymmetry, proportionately incorrect, grainy, noisy, overexposed, underexposed, chromatic aberration, lens flare, motion blur, glitch, color bleed, over-sharpened, washed out, watermark, time stamp, date stamp, 3D, 3D render, CGI, low-poly, pixelated, pixel art, voxel, cartoon, child-like drawing, crayon, graffiti, mosaic, collage, monotone, duotone, text focus, speech bubble, thought bubble, subtitles, captions, frames, borders, multiple panels, barcode, price tag, product label, watermark signature, cluttered background, distracting background, low-quality background, irrelevant objects, random objects, out-of-place elements, poorly drawn background, distorted background, over-saturated, under-saturated, color mismatch, color distortion, incorrect colors, unnatural colors, harsh lighting, flat lighting, low contrast, high contrast, shadow errors, lighting glitches, distorted perspective, impossible geometry, warped image, flattened image, incorrect depth of field, MC Escher, inconsistent art style, mixed media, mirror image, reflections, duplicates, clone stamp, repeating patterns, unintentional transparency, blended objects, unintentional merging, nsfw, nudity, explicit content, gore, violence, blood, injury, body horror, nightmare fuel 

• Anatomy and Proportions: AI models sometimes generate images with anatomical inconsistencies. These tags help eliminate images with distorted or incorrect anatomy.

asymmetrical eyes, distorted face, deformed face, incorrect facial features, bad face, extra eyes, extra mouth, missing eyes, missing mouth, mismatched facial features, unnatural pose, awkward pose, dislocated limbs, stretched limbs, shrunken limbs, floating limbs, disconnected limbs, distorted limbs, asymmetry, proportionately incorrect

• Image Quality and Artifacts: Tags related to image quality ensure the generated images are clean and free from technical imperfections like noise or glitches.

grainy, noisy, overexposed, underexposed, chromatic aberration, lens flare, motion blur, glitch, color bleed, over-sharpened, washed out, watermark, time stamp, date stamp

• Unwanted Styles and Mediums: If you're aiming for a specific art style, these tags help exclude other styles that might not fit your vision.

3D, 3D render, CGI, low-poly, pixelated, pixel art, voxel, cartoon, child-like drawing, crayon, graffiti, mosaic, collage, monotone, duotone

• Unwanted Elements: Removing distracting elements like text or watermarks keeps the focus on the main subject of the image.

text focus, speech bubble, thought bubble, subtitles, captions, frames, borders, multiple panels, barcode, price tag, product label, watermark signature

• Background and Environment Issues: Ensuring the background doesn't detract from the main subject enhances the overall image quality.

cluttered background, distracting background, low-quality background, irrelevant objects, random objects, out-of-place elements, poorly drawn background, distorted background

• Color and Lighting Issues: Proper lighting and color balance are crucial for realistic and appealing images.

over-saturated, under-saturated, color mismatch, color distortion, incorrect colors, unnatural colors, harsh lighting, flat lighting, low contrast, high contrast, shadow errors, lighting glitches

• Perspective and Depth Errors: Correct perspective ensures that the image looks natural and believable.

distorted perspective, impossible geometry, warped image, flattened image, incorrect depth of field, MC Escher

• Other Common Errors: Eliminating inconsistencies and unintended artifacts improves the cohesiveness of the image.

inconsistent art style, mixed media, mirror image, reflections, duplicates, clone stamp, repeating patterns, unintentional transparency, blended objects, unintentional merging

• Unwanted Content Types: Filtering out specific content types ensures the images align with the desired themes and comply with any content guidelines.

nsfw, nudity, explicit content, gore, violence, blood, injury, body horror, nightmare fuel 

Testing

A total of 3 prompts will be tested - full (main + extra), main and short.

Full

worst quality, low quality, bad quality, lowres, low res, pixelated, blurry, blurred, compression artifacts, jpeg artifacts, downscaled, what, bad anatomy, bad hands, worst hands, deformed hands, deformed fingers, deformed feet, deformed toes, extra limbs, extra arms, extra legs, extra legs and arms, extra fingers, extra digits, extra digit, mutated hands, mutation, mutated, fused fingers, fused toes, missing limbs, missing arms, missing fingers, missing toes, malformed, disfigured, wrong hands, ugly hands, ugly fingers, twisted hands, flexible deformity, conjoined, disembodied, text, watermark, signature, logo, url, username, tagme, bad, ugly, worst, very displeasing, displeasing, error, doesnotexist, unfinished, poorly drawn face, poorly drawn hands, poorly drawn feet, artistic error, bad proportions, bad perspective, bad foreshortening, bad twitter id, censored, bar censor, mosaic censoring, out of frame, cropped, missing, missing id, bad id, ai-generated, ai-assisted, stable diffusion, overly saturated, overly vivid, cross-eye, expressionless, extra, fewer, early, old, oldest, scan, sketch, manga, comic, monochrome, simple background, abstract, sequence, lineup, turnaround, 2koma, 4koma, microsoft paint \(medium\), artifacts, adversarial noise, has bad revision, scan artifacts, stitched, resized, image sample, asymmetrical eyes, distorted face, deformed face, incorrect facial features, bad face, extra eyes, extra mouth, missing eyes, missing mouth, mismatched facial features, unnatural pose, awkward pose, dislocated limbs, stretched limbs, shrunken limbs, floating limbs, disconnected limbs, distorted limbs, asymmetry, proportionately incorrect, grainy, noisy, overexposed, underexposed, chromatic aberration, lens flare, motion blur, glitch, color bleed, over-sharpened, washed out, watermark, time stamp, date stamp, 3D, 3D render, CGI, low-poly, pixelated, pixel art, voxel, cartoon, child-like drawing, crayon, graffiti, mosaic, collage, monotone, duotone, text focus, speech bubble, thought bubble, subtitles, captions, frames, borders, multiple panels, barcode, price tag, product label, watermark signature, cluttered background, distracting background, low-quality background, irrelevant objects, random objects, out-of-place elements, poorly drawn background, distorted background, over-saturated, under-saturated, color mismatch, color distortion, incorrect colors, unnatural colors, harsh lighting, flat lighting, low contrast, high contrast, shadow errors, lighting glitches, distorted perspective, impossible geometry, warped image, flattened image, incorrect depth of field, MC Escher, inconsistent art style, mixed media, mirror image, reflections, duplicates, clone stamp, repeating patterns, unintentional transparency, blended objects, unintentional merging, gore, violence, blood, injury, body horror, nightmare fuel 

Main

worst quality, low quality, bad quality, lowres, low res, pixelated, blurry, blurred, compression artifacts, jpeg artifacts, downscaled, what, bad anatomy, bad hands, worst hands, deformed hands, deformed fingers, deformed feet, deformed toes, extra limbs, extra arms, extra legs, extra legs and arms, extra fingers, extra digits, extra digit, mutated hands, mutation, mutated, fused fingers, fused toes, missing limbs, missing arms, missing fingers, missing toes, malformed, disfigured, wrong hands, ugly hands, ugly fingers, twisted hands, flexible deformity, conjoined, disembodied, text, watermark, signature, logo, url, username, tagme, bad, ugly, worst, very displeasing, displeasing, error, doesnotexist, unfinished, poorly drawn face, poorly drawn hands, poorly drawn feet, artistic error, bad proportions, bad perspective, bad foreshortening, bad twitter id, censored, bar censor, mosaic censoring, out of frame, cropped, missing, missing id, bad id, ai-generated, ai-assisted, stable diffusion, overly saturated, overly vivid, cross-eye, expressionless, extra, fewer, early, old, oldest, scan, sketch, manga, comic, monochrome, simple background, abstract, sequence, lineup, turnaround, 2koma, 4koma, microsoft paint \(medium\), artifacts, adversarial noise, has bad revision, scan artifacts, stitched, resized, image sample, 

Short

worst quality, low quality, bad quality, lowres, low res, pixelated, blurry, blurred, compression artifacts, jpeg artifacts, 

Full 1

Anime Aesthetic Score: 0.205539

Main 1

Anime Aesthetic Score: 0.336160

Short 1

Anime Aesthetic Score: 0.236479

Full 2

Anime Aesthetic Score: 0.818376

Main 2

Anime Aesthetic Score: 0.804808

Short 2

Anime Aesthetic Score: 0.669585

Full 3

Anime Aesthetic Score: 0.881179

Main 3

Anime Aesthetic Score: 0.879950

Short 3

Anime Aesthetic Score: 0.884813

Results:

While initial results varied, with some anomalies likely caused by the overly exhaustive negative prompt "Full", both "Full" and "Main" prompt consistently improved scores, achieving high aesthetic ratings in subsequent tests. Interestingly, the "Short" prompt also yielded a high score in the third test, suggesting that even basic quality negatives can be effective.