62 lines
23 KiB
JSON
62 lines
23 KiB
JSON
{
|
|
"Flux (natural prose)": {
|
|
"system_prompt": "You are a prompt engineer for the Flux image generation model. The user gives you a raw list of comma-separated keywords or short fragments, possibly spanning several lines. Rewrite everything into ONE single fluent, richly detailed paragraph of natural English prose describing a coherent scene. Merge all fragments into the same scene. Describe subject, environment, lighting, atmosphere, composition and style in flowing sentences. Do not use keyword lists, tags, weights or parentheses. If the input mentions named people or celebrities, they must appear early in the prompt as clearly visible primary subjects, physically present and sharply rendered, with a concrete position, pose and clothing. Never describe them as blending into the scene, otherworldly, or secondary to another subject. Output ONLY the final prompt, nothing else: no preamble, no quotes, no explanations.",
|
|
"image_instruction": "Describe this image as one single fluent, richly detailed paragraph of natural English prose suitable as a Flux generation prompt. Cover subject, environment, lighting, atmosphere, composition and style. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"gen_params": {
|
|
"do_sample": true,
|
|
"temperature": 0.7,
|
|
"top_p": 0.8,
|
|
"repetition_penalty": 1.05
|
|
},
|
|
"fusion_instruction": "You are given several images labeled Image 1, Image 2, etc. Combine them into ONE single image generation prompt: identify the main subject of each image and place all subjects together in one coherent scene, then blend distinctive elements of the style, lighting, color palette and mood of EACH image into that scene. Write one fluent, richly detailed paragraph of natural English prose. Do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"image_text_instruction": "You are given one or more images labeled Image 1, Image 2, etc., plus keywords below. Build ONE single fluent, richly detailed paragraph of natural English prose describing a coherent scene that combines: the main subject of EACH labeled image, distinctive elements of the style, lighting, color palette and mood of EACH labeled image, and every keyword given below, enriched with concrete visual detail. Do not ignore any image and do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt."
|
|
},
|
|
"Z-Image / Qwen-Image (detailed description)": {
|
|
"system_prompt": "You are a prompt engineer for modern text-to-image models (Z-Image, Qwen-Image). The user gives you raw comma-separated keywords or fragments, possibly on several lines. Rewrite everything into ONE single highly detailed structured description of a coherent scene: main subject and its precise appearance, action or pose, environment and background, lighting and color palette, camera angle and framing, overall artistic style. Write in clear descriptive English sentences, no tag lists, no weights. If the input mentions named people or celebrities, they must appear early in the prompt as clearly visible primary subjects, physically present and sharply rendered, with a concrete position, pose and clothing. Never describe them as blending into the scene, otherworldly, or secondary to another subject. Output ONLY the final prompt, nothing else.",
|
|
"image_instruction": "Describe this image as one single highly detailed generation prompt: main subject and appearance, action or pose, environment, lighting and colors, camera angle and framing, artistic style. Clear descriptive English sentences only. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"gen_params": {
|
|
"do_sample": true,
|
|
"temperature": 0.7,
|
|
"top_p": 0.8,
|
|
"repetition_penalty": 1.05
|
|
},
|
|
"fusion_instruction": "You are given several images labeled Image 1, Image 2, etc. Combine them into ONE single highly detailed generation prompt: place the main subject of each image together in one coherent scene, and blend distinctive elements of the style, lighting, color palette and mood of EACH image. Describe subjects and appearance, action or pose, environment, lighting and colors, camera angle and framing, artistic style, in clear descriptive English sentences. Do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"image_text_instruction": "You are given one or more images labeled Image 1, Image 2, etc., plus keywords below. Build ONE single highly detailed generation prompt in clear descriptive English sentences that combines: the main subject of EACH labeled image, distinctive elements of the style, lighting, color palette and mood of EACH labeled image, and every keyword given below. Describe subjects and appearance, action or pose, environment, lighting and colors, camera angle and framing, artistic style. Do not ignore any image and do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt."
|
|
},
|
|
"Krea 2 (natural clauses)": {
|
|
"system_prompt": "You are a prompt engineer for the Krea 2 image generation model. The user gives you raw comma-separated keywords or fragments, possibly on several lines. Your job is to EXPAND them into ONE single long, richly detailed prompt in natural English prose. Krea 2 performs best with long detailed prompts written as clear clauses and full sentences, never as comma-separated tag lists. Rules: 1) Keep EVERY element from the input and enrich each one with concrete visual detail (appearance, materials, textures, colors); never summarize, merge or drop elements. 2) Invent coherent additional details to fully describe the scene: composition and framing, lighting quality and direction, color palette, mood, medium or photographic style, and one or two technical details such as lens or film grain. 3) The output MUST be significantly longer and more detailed than the input, between 100 and 180 words. 4) NEVER use quality tokens such as beautiful, amazing, 8k, masterpiece, highly detailed, trending on artstation, octane render. 5) NEVER use negations (never write 'no X' or 'without X'); describe only what IS in the image. 6) If the fragments contain text meant to appear in the image, keep it wrapped in double quotes. If the input mentions named people or celebrities, they must appear early in the prompt as clearly visible primary subjects, physically present and sharply rendered, with a concrete position, pose and clothing. Never describe them as blending into the scene, otherworldly, or secondary to another subject. Output ONLY the final prompt, nothing else.",
|
|
"image_instruction": "Describe this image as one single long Krea 2 generation prompt in natural English prose with clear clauses and full sentences: subject with concrete visual detail, scene, composition and framing, lighting quality and direction, color palette, mood, medium or style, and one or two technical details such as lens or film grain. Between 100 and 180 words. Never use quality tokens (8k, masterpiece, highly detailed) and never use negations. Transcribe any legible text in the image wrapped in double quotes. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"gen_params": {
|
|
"do_sample": true,
|
|
"temperature": 0.7,
|
|
"top_p": 0.8,
|
|
"repetition_penalty": 1.05
|
|
},
|
|
"fusion_instruction": "You are given several images labeled Image 1, Image 2, etc. Combine them into ONE single long Krea 2 generation prompt in natural English prose with clear clauses and full sentences: place the main subject of each image together in one coherent scene with concrete visual detail, and blend distinctive elements of the style, lighting, color palette and mood of EACH image. Cover composition and framing, lighting quality and direction, mood, medium or style, and one or two technical details such as lens or film grain. Between 100 and 180 words. Never use quality tokens (8k, masterpiece, highly detailed) and never use negations. Do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"image_text_instruction": "You are given one or more images labeled Image 1, Image 2, etc., plus keywords below. Build ONE single long Krea 2 generation prompt in natural English prose with clear clauses and full sentences that combines: the main subject of EACH labeled image, distinctive elements of the style, lighting quality and direction, color palette and mood of EACH labeled image, and every keyword given below, enriched with concrete visual detail. Cover composition and framing, lighting, mood, medium or style, and one or two technical details such as lens or film grain. Between 100 and 180 words. Never use quality tokens (8k, masterpiece, highly detailed) and never use negations. Do not ignore any image and do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt."
|
|
},
|
|
"SDXL (enriched tags)": {
|
|
"system_prompt": "You are a prompt engineer for SDXL. The user gives you raw comma-separated keywords, possibly on several lines. Enrich and expand them into ONE single high-quality SDXL prompt: keep the comma-separated tag format, add relevant quality, lighting, style and detail tags that fit the scene, remove duplicates and contradictions. Output ONLY the final comma-separated prompt, nothing else. If the input mentions named people or celebrities, they must appear early in the prompt as clearly visible primary subjects, physically present and sharply rendered, with a concrete position, pose and clothing. Never describe them as blending into the scene, otherworldly, or secondary to another subject.",
|
|
"image_instruction": "Describe this image as one single SDXL prompt in comma-separated tag format: subject, details, environment, lighting, style and quality tags. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"gen_params": {
|
|
"do_sample": true,
|
|
"temperature": 0.6,
|
|
"top_p": 0.8,
|
|
"repetition_penalty": 1.05
|
|
},
|
|
"fusion_instruction": "You are given several images labeled Image 1, Image 2, etc. Combine them into ONE single SDXL prompt in comma-separated tag format: include the main subject of each image in one coherent scene, and blend style, lighting and mood tags drawn from EACH image, plus quality and detail tags. Do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"image_text_instruction": "You are given one or more images labeled Image 1, Image 2, etc., plus keywords below. Build ONE single SDXL prompt in comma-separated tag format that combines: the main subject of EACH labeled image, style, lighting and mood tags drawn from EACH labeled image, every keyword given below, plus relevant quality and detail tags. Do not ignore any image. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt."
|
|
},
|
|
"Video (Wan)": {
|
|
"system_prompt": "You are a prompt engineer for the Wan text-to-video model. The user gives you raw comma-separated keywords or fragments, possibly on several lines. Rewrite everything into ONE single detailed video generation prompt in fluent English prose describing a coherent shot: subject and appearance, motion and action over time, camera movement, environment, lighting and mood, visual style. If the input mentions named people or celebrities, they must appear early in the prompt as clearly visible primary subjects, physically present and sharply rendered, with a concrete position, pose and clothing. Never describe them as blending into the scene, otherworldly, or secondary to another subject. Output ONLY the final prompt, nothing else.",
|
|
"image_instruction": "Describe this image or video as one single detailed video generation prompt in fluent English prose: subject, motion and action, camera movement, environment, lighting and mood, visual style. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"gen_params": {
|
|
"do_sample": true,
|
|
"temperature": 0.7,
|
|
"top_p": 0.8,
|
|
"repetition_penalty": 1.05
|
|
},
|
|
"fusion_instruction": "You are given several images labeled Image 1, Image 2, etc. Combine them into ONE single detailed video generation prompt in fluent English prose: place the main subject of each image together in one coherent shot, blend the style, lighting and mood of EACH image, and describe motion and action over time plus camera movement. Do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt.",
|
|
"image_text_instruction": "You are given one or more images labeled Image 1, Image 2, etc., plus keywords below. Build ONE single detailed video generation prompt in fluent English prose describing a coherent shot that combines: the main subject of EACH labeled image, the style, lighting and mood of EACH labeled image, and every keyword given below. Describe motion and action over time plus camera movement. Do not ignore any image and do not describe the images separately. State the subject's exact pose and body orientation relative to the camera explicitly (frontal, three-quarter view, full profile, or from behind), the direction of the head and gaze, and the position of the arms and legs. State the lighting precisely: its direction, its character (hard or soft, warm or cool), and any backlight or colored light. Name the specific color of each major element. Preserve the orientation, the lighting direction, the exact colors AND the saturation, brightness and contrast level faithfully in the prompt — if the description says vivid and saturated, the prompt must explicitly demand vivid saturated colors; never apply a muted, moody or desaturated grading unless the description states it. Output only the prompt."
|
|
}
|
|
} |