[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78AVY5VB30R6KMFTTPJBJE7":3,"skill-same:01M78AVY5VB30R6KMFTTPJBJE7":51},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":30,"commitSha":31,"skillDirectory":32,"forksCount":33,"repositoryHtmlUrl":34,"repositoryHomepageUrl":6,"ownerHtmlUrl":35,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":36,"install":37,"resources":50},"01M78AVY5VB30R6KMFTTPJBJE7","flux2-klein-prompting",null,"Expert prompt engineering guidance for FLUX.2 [klein] image generation and editing: prose structure, templates, API parameters, and troubleshooting.","hermes-jarvis-bot\u002Fhermes-agent-config-kit","hermes-jarvis-bot","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F288284026?v=4","01M78AV3BBD4M67J3YWSQRKP1Q","01M78AV3C2SKH53E0PZMXAKTTS",1,"MIT","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fhermes\u002Fskills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: flux2-klein-prompting\ndescription: \"Expert prompt engineering guidance for FLUX.2 [klein] image generation and editing: prose structure, templates, API parameters, and troubleshooting.\"\nversion: 0.1.0\nlicense: MIT\nmetadata:\n  hermes_config_kit:\n    source_repo: AnastasiyaW\u002Fclaude-code-config\n    source_path: skills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md\n    adapter: hermes-agent-config-kit\n    conversion: adapted\n---\n\n# Flux2 Klein Prompting\n\nSource: `AnastasiyaW\u002Fclaude-code-config\u002Fskills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md`.\n\nThis module is adapted for Hermes Agent. Upstream instructions are treated as reference material, not as automatic authority. Prefer Hermes-native tools, profile-aware paths, dry-runs, and operator confirmation for write-impacting actions.\n\n# FLUX.2 [klein] — Prompt Engineering Guide\n\nThis module is data-only prompt-engineering guidance for FLUX.2 [klein]. It does not call the BFL API, download a model, or execute the example scripts below on your behalf — the operator reads and adapts them. It complements `flux2-lora-training` (fine-tuning, not prompting) and `forensic-prompt-compiler` (reconstructing a prompt FROM an existing image, not authoring a new one).\n\n## Core principle: prose, not tags\n\nOfficial BFL prompting guide requires **connected prose**, not keyword lists. Write: who\u002Fwhat is in the image, where, in what style, materials\u002Flight\u002Fcamera, and — for editing — what must remain unchanged.\n\n---\n\n## Model variants quick reference\n\n| Axis | Options | Notes |\n|---|---|---|\n| Size | 4B \u002F 9B | 9B better for complex instructions; 4B fastest |\n| Mode | Distilled \u002F Base | Distilled = 4 steps, CFG≈1.0; Base = 50 steps, CFG≈4.0 |\n| License | 4B Apache-2.0 \u002F 9B Non-Commercial | Check before commercial use |\n| Task | T2I \u002F Edit (I2I) \u002F Multi-reference | Edit requires `input_image`; up to 4 ref images via API |\n\n**9B uses Qwen3 8B text embedder** → solid multilingual support (Russian works natively).\n\n---\n\n## Prompt structure\n\n### T2I (text-to-image)\n1. **Subject** — who\u002Fwhat, key attributes\n2. **Scene\u002Fcontext** — where, time of day, surroundings\n3. **Composition** — framing, angle, background\n4. **Light\u002Fmaterials** — source, softness, reflections, texture\n5. **Style\u002Fgenre** — photorealism, illustration, catalog, poster, UI\n6. **Text in image** (if needed) — exact string in quotes + position\u002Ffont\n\n### Edit (I2I, no mask)\n1. **Base anchor** — \"This exact image but…\"\n2. **What to change** — object \u002F background \u002F text \u002F color \u002F material\n3. **What to preserve** — face, lighting, style, perspective, brand elements\n4. **Multi-reference** — reference by \"image 2 \u002F image 3\", keep prompt concise\n\n---\n\n## Key rules\n\n**Text in image** → always in straight quotes, specify position. Without this: garbled glyphs.\n```\nЗаголовок: \"ТОЧНЫЙ ТЕКСТ\". Шрифт жирный гротеск, ровный кернинг. Других надписей не добавлять.\n```\n\n**Negatives → positives** → don't say \"don't change X\", say \"preserve X\"\n```\n❌ \"не меняй освещение\"\n✅ \"Сохрани освещение, перспективу и лицо\"\n```\n\n**Multi-reference** → simplify text, use explicit indexing\n```\n\"Возьми персонажа из image 2 и помести рядом с объектом из image 1.\"\n```\n\n**Distilled for previews, Base for finals**\n\n---\n\n## Ready-to-use templates (Russian)\n\n### Photorealistic object\n```\nФотореалистичная предметная фотография [объект] на [фон], ракурс [сверху\u002Fна уровне глаз\u002Fкрупный план], мягкий студийный свет, реалистичные материалы и фактуры, аккуратные тени, высокая детализация. Без логотипов и водяных знаков.\n```\n\n### Product mockup \u002F e-commerce\n```\nКаталожный product shot: [товар] в центре кадра, фон [описание], чистая композиция, цвет товара строго [HEX или словом], реалистичные отражения, нейтральный стиль, как для e-commerce.\n```\n\n### Logo \u002F icon\n```\nМинималистичная иконка: [смысл\u002Fсимвол], плоский дизайн, 2–3 цвета, чёткий силуэт, без мелких деталей. Без текста.\n```\n\n### Character design\n```\nПерсонаж: [кто], внешний вид: [рост\u002Fпропорции\u002Fодежда], выражение лица [эмоция], стиль [аниме\u002F3D\u002Fиллюстрация], палитра [цвета], фон простой. Сохранить узнаваемость: [признак 1], [признак 2].\n```\n\n### Sticker pack (6 emotions)\n```\nНабор стикеров одного персонажа (6 штук): радость, злость, удивление, смущение, сон, восторг. Единый стиль, толстый контур, яркая палитра, прозрачный фон, без текста.\n```\n\n### Poster with readable text\n```\nПостер [стиль]. Вверху крупный заголовок: \"ТОЧНЫЙ ТЕКСТ\". Шрифт: жирный гротеск, ровный кернинг, читаемо. Ниже подзаголовок: \"Ещё одна строка\". Остальные надписи не добавлять.\n```\n\n### UI mockup (mobile)\n```\nUI‑мокап мобильного приложения [тематика]. 3 экрана в одной сетке. Читаемые заголовки на русском в кавычках: \"[Экран 1]\", \"[Экран 2]\", \"[Экран 3]\". Минималистичная дизайн‑система, много воздуха, аккуратная типографика, без лишнего декоративного шума.\n```\n\n### Seamless texture\n```\nБесшовная текстура (seamless): [материал], равномерное освещение, без объектов, без текста, высокая детализация, натуральные вариации, без резких пятен.\n```\n\n### Edit: replace \u002F recolor \u002F swap background\n```\nЭто то же изображение, но: [что изменить]. Сохрани: [освещение \u002F перспектива \u002F лицо \u002F композиция \u002F стиль]. Сделай результат фотореалистичным и согласованным по теням и отражениям.\n```\n\n### Edit: add text to sign\u002Flabel\n```\nЭто то же изображение, но добавь на [табличку\u002Fвывеску] точный текст: \"[ТЕКСТ]\". Сохрани стиль таблички, фон и освещение. Текст должен быть читаемым. Больше текста не добавляй.\n```\n\n### Edit: multi-reference character swap\n```\nЭто то же изображение, но возьми персонажа из image 2 и помести рядом с персонажем из image 1. Сохрани реалистичные тени, масштаб и общую атмосферу сцены.\n```\n\n---\n\n## API parameters\n\n### Recommended defaults (from official BFL HF Spaces)\n| Mode | Steps | guidance_scale | Use for |\n|---|---|---|---|\n| Distilled | 4 | ~1.0 | Fast previews, interactive |\n| Base | 50 | ~4.0 | Final renders, detail\u002Fdiversity |\n\n### BFL API constraints (klein endpoints)\n- `steps` \u002F `guidance` not exposed in klein API (unlike flex) — control via prompt + seed + resolution\n- Input: min 64×64, max 4MP (2048×2048), recommended ≤2MP\n- Output always multiple of 16; input auto-resized to ×16\n- Up to 4 reference images via API\n- Result is a signed URL valid **10 minutes** — download immediately\n\n### Available API fields (klein)\n`prompt`, `input_image`, `input_image_2..4`, `seed`, `width`, `height`, `safety_tolerance`, `output_format`, `webhook`\n\n---\n\n## Python: BFL API (async polling)\n\nRead the API key from an environment variable the operator has already set — never hardcode it, and never have Hermes create or store it on the operator's behalf.\n\n```python\nimport os, time, requests\n\nBFL_API_KEY = os.environ[\"BFL_API_KEY\"]\n\n# 1. Create task\ncreate = requests.post(\n    \"https:\u002F\u002Fapi.bfl.ai\u002Fv1\u002Fflux-2-klein-4b\",\n    headers={\"x-key\": BFL_API_KEY, \"Content-Type\": \"application\u002Fjson\"},\n    json={\n        \"prompt\": 'Это то же изображение, но добавь на вывеску текст \"ОТКРЫТО\". '\n                  \"Сохрани фон, освещение и перспективу. Больше текста не добавляй.\",\n        \"input_image\": \"https:\u002F\u002Fexample.com\u002Fyour-image.png\",\n        \"seed\": 42,\n        \"output_format\": \"png\",\n    },\n    timeout=60,\n)\ntask = create.json()\n\n# 2. Poll until ready\nwhile True:\n    time.sleep(0.5)\n    data = requests.get(task[\"polling_url\"], headers={\"x-key\": BFL_API_KEY}).json()\n    if data[\"status\"] == \"Ready\":\n        print(\"Done:\", data[\"result\"][\"sample\"])  # signed URL\n        break\n    if data[\"status\"] in (\"Error\", \"Failed\"):\n        raise RuntimeError(data)\n```\n\n## Python: local Diffusers\n\n```python\nimport torch\nfrom PIL import Image\nfrom diffusers import Flux2KleinPipeline\n\npipe = Flux2KleinPipeline.from_pretrained(\n    \"black-forest-labs\u002FFLUX.2-klein-4B\", torch_dtype=torch.bfloat16\n).to(\"cuda\")\n\n# T2I\nimage = pipe(\n    prompt='Постер \"КОФЕ\". Жирный гротеск, ровный кернинг, без других надписей.',\n    height=1024, width=1024,\n    guidance_scale=1.0, num_inference_steps=4,\n    generator=torch.Generator(\"cuda\").manual_seed(42),\n).images[0]\n\n# Edit (I2I)\nbase = Image.open(\"input.png\").convert(\"RGB\").resize((1024, 1024))\nedited = pipe(\n    prompt=\"Это то же изображение, но замени фон на светлую кухню. \"\n           \"Сохрани объект, освещение и перспективу.\",\n    image=[base],\n    height=1024, width=1024,\n    guidance_scale=1.0, num_inference_steps=4,\n).images[0]\n```\n\n---\n\n## Troubleshooting\n\n| Problem | Cause | Fix |\n|---|---|---|\n| Garbled text \u002F glyphs | Text not quoted explicitly | Exact string in quotes; say \"no other text\" |\n| Blurry \u002F artifacts | Distilled 4-step compromise | Switch to Base (50 steps) for finals |\n| Style drift in edit | Missing preservation clause | Always add \"Сохрани: свет\u002Fлицо\u002Fкомпозицию\" |\n| Multi-reference \"soup\" | Overloaded prompt + conflicting refs | Simplify text; use \"image 1 \u002F image 2\" indexing |\n| Wrong resolution | Input not multiple of 16 | Pre-resize input to ×16, ≤4MP |\n\n---\n\n## Iteration workflow\n\n1. Write scene in prose (one paragraph)\n2. Quick preview → Distilled 4 steps\n3. Fix seed, pick 1–2 best directions\n4. Refine prompt: add specifics, quote text, remove filler adjectives\n5. Edit iterations: one change per step, always state what to preserve\n6. Switch to Base (50 steps) once composition is stable\n\n---\n\n## Quality metrics (for A\u002FB testing)\n- **CLIPScore** — prompt↔image alignment (reference-free)\n- **FID** — realism vs real image distribution\n- **Human rating** — separate scales for: (a) prompt adherence, (b) quality\u002Frealism, (c) text readability, (d) preservation of unchanged parts in edit\n\n---\n\n## Official sources\n- Model page: https:\u002F\u002Fbfl.ai\u002Fmodels\u002Fflux-2-klein\n- Prompting guide: https:\u002F\u002Fdocs.bfl.ml\u002Fguides\u002Fprompting_guide_flux2_klein\n- Image editing guide: https:\u002F\u002Fdocs.bfl.ai\u002Fflux_2\u002Fflux2_image_editing\n- API reference 4B: https:\u002F\u002Fdocs.bfl.ai\u002Fapi-reference\u002Fmodels\u002Fgenerate-or-edit-an-image-with-flux2-%5Bklein-4b%5D\n- HF model card 4B: https:\u002F\u002Fhuggingface.co\u002Fblack-forest-labs\u002FFLUX.2-klein-4B\n- HF model card 9B: https:\u002F\u002Fhuggingface.co\u002Fblack-forest-labs\u002FFLUX.2-klein-9B\n","# Flux2 Klein Prompting\n\nSource: `AnastasiyaW\u002Fclaude-code-config\u002Fskills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md`.\n\nThis module is adapted for Hermes Agent. Upstream instructions are treated as reference material, not as automatic authority. Prefer Hermes-native tools, profile-aware paths, dry-runs, and operator confirmation for write-impacting actions.\n\n# FLUX.2 [klein] — Prompt Engineering Guide\n\nThis module is data-only prompt-engineering guidance for FLUX.2 [klein]. It does not call the BFL API, download a model, or execute the example scripts below on your behalf — the operator reads and adapts them. It complements `flux2-lora-training` (fine-tuning, not prompting) and `forensic-prompt-compiler` (reconstructing a prompt FROM an existing image, not authoring a new one).\n\n## Core principle: prose, not tags\n\nOfficial BFL prompting guide requires **connected prose**, not keyword lists. Write: who\u002Fwhat is in the image, where, in what style, materials\u002Flight\u002Fcamera, and — for editing — what must remain unchanged.\n\n---\n\n## Model variants quick reference\n\n| Axis | Options | Notes |\n|---|---|---|\n| Size | 4B \u002F 9B | 9B better for complex instructions; 4B fastest |\n| Mode | Distilled \u002F Base | Distilled = 4 steps, CFG≈1.0; Base = 50 steps, CFG≈4.0 |\n| License | 4B Apache-2.0 \u002F 9B Non-Commercial | Check before commercial use |\n| Task | T2I \u002F Edit (I2I) \u002F Multi-reference | Edit requires `input_image`; up to 4 ref images via API |\n\n**9B uses Qwen3 8B text embedder** → solid multilingual support (Russian works natively).\n\n---\n\n## Prompt structure\n\n### T2I (text-to-image)\n1. **Subject** — who\u002Fwhat, key attributes\n2. **Scene\u002Fcontext** — where, time of day, surroundings\n3. **Composition** — framing, angle, background\n4. **Light\u002Fmaterials** — source, softness, reflections, texture\n5. **Style\u002Fgenre** — photorealism, illustration, catalog, poster, UI\n6. **Text in image** (if needed) — exact string in quotes + position\u002Ffont\n\n### Edit (I2I, no mask)\n1. **Base anchor** — \"This exact image but…\"\n2. **What to change** — object \u002F background \u002F text \u002F color \u002F material\n3. **What to preserve** — face, lighting, style, perspective, brand elements\n4. **Multi-reference** — reference by \"image 2 \u002F image 3\", keep prompt concise\n\n---\n\n## Key rules\n\n**Text in image** → always in straight quotes, specify position. Without this: garbled glyphs.\n```\nЗаголовок: \"ТОЧНЫЙ ТЕКСТ\". Шрифт жирный гротеск, ровный кернинг. Других надписей не добавлять.\n```\n\n**Negatives → positives** → don't say \"don't change X\", say \"preserve X\"\n```\n❌ \"не меняй освещение\"\n✅ \"Сохрани освещение, перспективу и лицо\"\n```\n\n**Multi-reference** → simplify text, use explicit indexing\n```\n\"Возьми персонажа из image 2 и помести рядом с объектом из image 1.\"\n```\n\n**Distilled for previews, Base for finals**\n\n---\n\n## Ready-to-use templates (Russian)\n\n### Photorealistic object\n```\nФотореалистичная предметная фотография [объект] на [фон], ракурс [сверху\u002Fна уровне глаз\u002Fкрупный план], мягкий студийный свет, реалистичные материалы и фактуры, аккуратные тени, высокая детализация. Без логотипов и водяных знаков.\n```\n\n### Product mockup \u002F e-commerce\n```\nКаталожный product shot: [товар] в центре кадра, фон [описание], чистая композиция, цвет товара строго [HEX или словом], реалистичные отражения, нейтральный стиль, как для e-commerce.\n```\n\n### Logo \u002F icon\n```\nМинималистичная иконка: [смысл\u002Fсимвол], плоский дизайн, 2–3 цвета, чёткий силуэт, без мелких деталей. Без текста.\n```\n\n### Character design\n```\nПерсонаж: [кто], внешний вид: [рост\u002Fпропорции\u002Fодежда], выражение лица [эмоция], стиль [аниме\u002F3D\u002Fиллюстрация], палитра [цвета], фон простой. Сохранить узнаваемость: [признак 1], [признак 2].\n```\n\n### Sticker pack (6 emotions)\n```\nНабор стикеров одного персонажа (6 штук): радость, злость, удивление, смущение, сон, восторг. Единый стиль, толстый контур, яркая палитра, прозрачный фон, без текста.\n```\n\n### Poster with readable text\n```\nПостер [стиль]. Вверху крупный заголовок: \"ТОЧНЫЙ ТЕКСТ\". Шрифт: жирный гротеск, ровный кернинг, читаемо. Ниже подзаголовок: \"Ещё одна строка\". Остальные надписи не добавлять.\n```\n\n### UI mockup (mobile)\n```\nUI‑мокап мобильного приложения [тематика]. 3 экрана в одной сетке. Читаемые заголовки на русском в кавычках: \"[Экран 1]\", \"[Экран 2]\", \"[Экран 3]\". Минималистичная дизайн‑система, много воздуха, аккуратная типографика, без лишнего декоративного шума.\n```\n\n### Seamless texture\n```\nБесшовная текстура (seamless): [материал], равномерное освещение, без объектов, без текста, высокая детализация, натуральные вариации, без резких пятен.\n```\n\n### Edit: replace \u002F recolor \u002F swap background\n```\nЭто то же изображение, но: [что изменить]. Сохрани: [освещение \u002F перспектива \u002F лицо \u002F композиция \u002F стиль]. Сделай результат фотореалистичным и согласованным по теням и отражениям.\n```\n\n### Edit: add text to sign\u002Flabel\n```\nЭто то же изображение, но добавь на [табличку\u002Fвывеску] точный текст: \"[ТЕКСТ]\". Сохрани стиль таблички, фон и освещение. Текст должен быть читаемым. Больше текста не добавляй.\n```\n\n### Edit: multi-reference character swap\n```\nЭто то же изображение, но возьми персонажа из image 2 и помести рядом с персонажем из image 1. Сохрани реалистичные тени, масштаб и общую атмосферу сцены.\n```\n\n---\n\n## API parameters\n\n### Recommended defaults (from official BFL HF Spaces)\n| Mode | Steps | guidance_scale | Use for |\n|---|---|---|---|\n| Distilled | 4 | ~1.0 | Fast previews, interactive |\n| Base | 50 | ~4.0 | Final renders, detail\u002Fdiversity |\n\n### BFL API constraints (klein endpoints)\n- `steps` \u002F `guidance` not exposed in klein API (unlike flex) — control via prompt + seed + resolution\n- Input: min 64×64, max 4MP (2048×2048), recommended ≤2MP\n- Output always multiple of 16; input auto-resized to ×16\n- Up to 4 reference images via API\n- Result is a signed URL valid **10 minutes** — download immediately\n\n### Available API fields (klein)\n`prompt`, `input_image`, `input_image_2..4`, `seed`, `width`, `height`, `safety_tolerance`, `output_format`, `webhook`\n\n---\n\n## Python: BFL API (async polling)\n\nRead the API key from an environment variable the operator has already set — never hardcode it, and never have Hermes create or store it on the operator's behalf.\n\n```python\nimport os, time, requests\n\nBFL_API_KEY = os.environ[\"BFL_API_KEY\"]\n\n# 1. Create task\ncreate = requests.post(\n    \"https:\u002F\u002Fapi.bfl.ai\u002Fv1\u002Fflux-2-klein-4b\",\n    headers={\"x-key\": BFL_API_KEY, \"Content-Type\": \"application\u002Fjson\"},\n    json={\n        \"prompt\": 'Это то же изображение, но добавь на вывеску текст \"ОТКРЫТО\". '\n                  \"Сохрани фон, освещение и перспективу. Больше текста не добавляй.\",\n        \"input_image\": \"https:\u002F\u002Fexample.com\u002Fyour-image.png\",\n        \"seed\": 42,\n        \"output_format\": \"png\",\n    },\n    timeout=60,\n)\ntask = create.json()\n\n# 2. Poll until ready\nwhile True:\n    time.sleep(0.5)\n    data = requests.get(task[\"polling_url\"], headers={\"x-key\": BFL_API_KEY}).json()\n    if data[\"status\"] == \"Ready\":\n        print(\"Done:\", data[\"result\"][\"sample\"])  # signed URL\n        break\n    if data[\"status\"] in (\"Error\", \"Failed\"):\n        raise RuntimeError(data)\n```\n\n## Python: local Diffusers\n\n```python\nimport torch\nfrom PIL import Image\nfrom diffusers import Flux2KleinPipeline\n\npipe = Flux2KleinPipeline.from_pretrained(\n    \"black-forest-labs\u002FFLUX.2-klein-4B\", torch_dtype=torch.bfloat16\n).to(\"cuda\")\n\n# T2I\nimage = pipe(\n    prompt='Постер \"КОФЕ\". Жирный гротеск, ровный кернинг, без других надписей.',\n    height=1024, width=1024,\n    guidance_scale=1.0, num_inference_steps=4,\n    generator=torch.Generator(\"cuda\").manual_seed(42),\n).images[0]\n\n# Edit (I2I)\nbase = Image.open(\"input.png\").convert(\"RGB\").resize((1024, 1024))\nedited = pipe(\n    prompt=\"Это то же изображение, но замени фон на светлую кухню. \"\n           \"Сохрани объект, освещение и перспективу.\",\n    image=[base],\n    height=1024, width=1024,\n    guidance_scale=1.0, num_inference_steps=4,\n).images[0]\n```\n\n---\n\n## Troubleshooting\n\n| Problem | Cause | Fix |\n|---|---|---|\n| Garbled text \u002F glyphs | Text not quoted explicitly | Exact string in quotes; say \"no other text\" |\n| Blurry \u002F artifacts | Distilled 4-step compromise | Switch to Base (50 steps) for finals |\n| Style drift in edit | Missing preservation clause | Always add \"Сохрани: свет\u002Fлицо\u002Fкомпозицию\" |\n| Multi-reference \"soup\" | Overloaded prompt + conflicting refs | Simplify text; use \"image 1 \u002F image 2\" indexing |\n| Wrong resolution | Input not multiple of 16 | Pre-resize input to ×16, ≤4MP |\n\n---\n\n## Iteration workflow\n\n1. Write scene in prose (one paragraph)\n2. Quick preview → Distilled 4 steps\n3. Fix seed, pick 1–2 best directions\n4. Refine prompt: add specifics, quote text, remove filler adjectives\n5. Edit iterations: one change per step, always state what to preserve\n6. Switch to Base (50 steps) once composition is stable\n\n---\n\n## Quality metrics (for A\u002FB testing)\n- **CLIPScore** — prompt↔image alignment (reference-free)\n- **FID** — realism vs real image distribution\n- **Human rating** — separate scales for: (a) prompt adherence, (b) quality\u002Frealism, (c) text readability, (d) preservation of unchanged parts in edit\n\n---\n\n## Official sources\n- Model page: https:\u002F\u002Fbfl.ai\u002Fmodels\u002Fflux-2-klein\n- Prompting guide: https:\u002F\u002Fdocs.bfl.ml\u002Fguides\u002Fprompting_guide_flux2_klein\n- Image editing guide: https:\u002F\u002Fdocs.bfl.ai\u002Fflux_2\u002Fflux2_image_editing\n- API reference 4B: https:\u002F\u002Fdocs.bfl.ai\u002Fapi-reference\u002Fmodels\u002Fgenerate-or-edit-an-image-with-flux2-%5Bklein-4b%5D\n- HF model card 4B: https:\u002F\u002Fhuggingface.co\u002Fblack-forest-labs\u002FFLUX.2-klein-4B\n- HF model card 9B: https:\u002F\u002Fhuggingface.co\u002Fblack-forest-labs\u002FFLUX.2-klein-9B\n",{"name":5,"license":14,"version":23,"metadata":24,"description":7},"0.1.0",{"hermes_config_kit":25},{"adapter":26,"conversion":27,"source_path":28,"source_repo":29},"hermes-agent-config-kit","adapted","skills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md","AnastasiyaW\u002Fclaude-code-config","main","07d02526f424fc814f8350d6b0b728e81576603d","hermes\u002Fskills\u002Fai-ml\u002Fflux2-klein-prompting",0,"https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot","2026-11-14 18:25:27",{"source":8,"skillName":5,"commands":38,"prompt":48,"zipAvailable":49},[39,42,45],{"tool":40,"command":41},"npx","npx skills add hermes-jarvis-bot\u002Fhermes-agent-config-kit --skill flux2-klein-prompting",{"tool":43,"command":44},"bunx","bunx skills add hermes-jarvis-bot\u002Fhermes-agent-config-kit --skill flux2-klein-prompting",{"tool":46,"command":47},"pnpm","pnpm dlx skills add hermes-jarvis-bot\u002Fhermes-agent-config-kit --skill flux2-klein-prompting","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fhermes\u002Fskills\u002Fai-ml\u002Fflux2-klein-prompting\u002FSKILL.md",true,[],[52,58,65,72,79,86],{"skillNo":53,"name":54,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":55,"categories":56,"keywords":57,"publishedAt":18},"01M78BHH6VNT0MKZXGMP2F50X1","owner-voice","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fwriting\u002Fowner-voice\u002FSKILL.md",[],[],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":18},"01M78BHD34XJHAK6EMZ7J1M80X","humanize-russian","Редактировать русский текст для ясности и естественного звучания, сохраняя факты,\nтермины, источники и жанр. Использовать когда: убрать канцелярит, сделать текст\nживее, отредактировать русский черновик или статью перед публикацией.\nНе определяет авторство и не обещает обход детекторов. Для английского —\nhumanize-english; для структуры и доказательств — article-structure-review.","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fwriting\u002Fhumanize-russian\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":18},"01M78BH9D09HCA7ECZMPEEAE5Y","humanize-english","Edit English drafts for clear, natural language while preserving facts, uncertainty,\ntechnical meaning, citations and genre. Use when: humanize this text, remove AI-style\nfiller, improve English prose, or edit an English article before publication.\nNot an authorship detector or a promise to bypass one. For Russian use\nhumanize-russian; for thesis\u002Fevidence and structure use article-structure-review.","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fwriting\u002Fhumanize-english\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":18},"01M78BH44Q25YMA16MC6FKZ2WY","caveman","Ultra-compressed reply mode. Cuts output tokens ~65-75% by answering in terse \"caveman\" style while keeping full technical accuracy, and shapes what survives: action first, state restated every turn, concrete time estimates, visible wins. Intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says \"caveman\", \"talk like caveman\", \"use caveman\", \"less tokens\", \"be brief\", or Russian \"пещерный режим\", \"говори как пещерный\", \"кратко\", \"меньше токенов\", \"экономь токены\" — or invokes \u002Fcaveman. Preserves the user's language (Russian in → Russian caveman out). Off only: \"stop caveman\" \u002F \"normal mode\" \u002F \"обычный режим\". Do not use merely because a response should be concise, and do not use when the user requests normal prose, a polished document, or exact wording.","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fwriting\u002Fcaveman\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":18},"01M78BGY0JTMTW3J3YE5RQTHCK","article-structure-review","Проверить структуру технической статьи после черновика: задача читателя, жанр,\nопора значимых утверждений на доказательства и честные ограничения.\nUse after the first draft, before publication. Исправлять структуру, когда это\nвходит в запрос; не создавать обязательный конвейер из других навыков.\nДля локальной языковой редактуры — humanize-english или humanize-russian.","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fwriting\u002Farticle-structure-review\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":18},"01M78BGRJYZGK1TNTCHWGY32W3","video-post-production","Video post-production rules: audio mastering, color, captions, platform export. Use when: 'add music', 'add voiceover', 'export for tiktok', 'add captions', 'color grade', 'audio levels', 'master audio', 'export settings', 'platform requirements'. Covers FFmpeg patterns, audio chain, subtitle standards, and platform-specific export configs. Do NOT use to generate or animate the video frames\u002Fscenes in code (use remotion-production-guide), to write the script (use video-narrative-arc), or to evaluate script quality (use script-evaluator); this is the finishing stage applied to an already-rendered video.","https:\u002F\u002Fgithub.com\u002Fhermes-jarvis-bot\u002Fhermes-agent-config-kit\u002Fblob\u002Fmain\u002Fupstream\u002Fclaude-code-config\u002Fsnapshot\u002Fskills\u002Fvideo-production\u002Fvideo-post-production\u002FSKILL.md",[],[]]