28건 · 관련도순
Add a diffusion model or Diffusers generation pipeline to SGLang. Examine reference code, choose a suitable pipeline structure, reuse existing components, and implement model settings and processing stages, with guidance for later multi-GPU support.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Run diffusers generation pipelines from a terminal and save their outputs. Inspect accepted inputs without downloading model weights, apply LoRA adapters during generation, run remotely through HF Jobs, and inspect software and GPU environment details.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Review a diffusers contribution before opening a pull request. Check the full change against project rules, trace potentially unused code, and receive a severity-based report separating required fixes from points to discuss with reviewers.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Integrate a model or pipeline into diffusers, from file structure and weight conversion to tests and documentation. Verify numerical agreement with the original implementation, with modular pipelines as the preferred starting point for new integrations.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Package an existing custom diffusers block into a directory ready for Hub upload. Select a class from a Python file and save its source, configuration, and associated files so others can load the block.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Build image generation workflows in Python with Stable Diffusion and HuggingFace Diffusers. The guide covers text prompts, transforming existing images, replacing selected regions, controlling composition, applying learned styles, and adjusting memory use during generation.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Build image generation workflows in Python with Stable Diffusion and HuggingFace Diffusers. The guide covers text prompts, transforming existing images, replacing selected regions, controlling composition, applying learned styles, and adjusting memory use during generation.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Create Python workflows for generating images from text, restyling existing pictures, and filling selected regions. Using HuggingFace Diffusers, the skill explains composition controls, style adapters, generation settings, and ways to reduce memory use.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Provides guidance for writing and benchmarking optimized Triton kernels for AMD GPUs (MI355X, R9700) on ROCm, targeting HuggingFace diffusers (LTX-Video, SD3, FLUX) and transformers. Core kernels: RMSNorm, RoPE 3D, GEGLU, AdaLN. Includes XCD swizzle, autotune, diffusers integration patterns, and LTX-Video pipeline injection.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use this skill for Hugging Face Diffusers tasks: pipeline inference, schedulers, adapters/loaders, training recipes, modular pipelines, conversion helpers, CLI checks, and repo maintenance.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use when the user wants to run a diffusers pipeline from a terminal (one-off generation, batch jobs, smoke-testing a new model), run on HF Sandbox hardware via `--remote`, introspect a pipeline's input schema before calling it, or attach a LoRA at inference time. Prefer this over writing ad-hoc Python scripts for generation tasks.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Geração de imagens com tecnologia de ponta usando Stable Diffusion via HuggingFace Diffusers. Use quando gerar imagens a partir de prompts em texto, realizar tradução de imagem para imagem, inpainting ou construir pipelines de difusão customizados.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Helps agents prepare image, video, speech, music, and avatar generation requests with an exact model and API host. It explains service-specific inputs, request validation, and job resumption so unavailable models are not silently replaced.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Measure denoising latency in SGLang diffusion models and identify slow operations with profiler traces. Compare performance and output quality on the same GPUs, check existing optimizations, and gather evidence before starting low-level kernel work.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Build and publish a browser demo for a supplied LoRA, a model adaptation for image or video tasks. Tailor the Gradio interface to its inputs and follow the model’s recommended settings on Hugging Face Spaces.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Provides guidance for writing and benchmarking optimized CUDA kernels for NVIDIA GPUs (H100, A100, T4) targeting HuggingFace diffusers and transformers libraries. Kernels must be kernel-builder/ABI3-compliant: no pybind11, no setup.py, TORCH_LIBRARY_EXPAND bindings only. Supports models like LTX-Video, Stable Diffusion, LLaMA, Mistral, and Qwen. Includes integration with HuggingFace Kernels Hub (get_kernel) for loading pre-compiled kernels. Includes benchmarking scripts to compare kernel performance against baseline implementations.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Route Custom Diffusion workflows for concept data prep, diffusers training and SDXL, sample generation, delta tools, and CustomConcept101 evaluation.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
AIGC 创作工作流 (AIGC 创作工作流 / AIGC Creative Workflow — 以「生成式 AI 视觉创作（图像 + 视频）的工程化创作流水线」为对象的认知操作系统：从「需求/概念（要做什么画面/镜头/风格）→ 模型与工具选型（Midjourney / Stable Diffusion·SDXL·Flux / ComfyUI / 可灵 Kling / 即梦 Jimeng / Runway / Vidu / Sora）→ Prompt 与参数设计（提示词工程、风格描述、负向、种子、CFG/步数/采样器、参考图）→ 可控生成（ControlNet / LoRA / IPAdapter / inpainting / 图生图 / 局部重绘 / 区域控制）→ 批量出图与筛选 → 精修与合成（放大/超分、修手修脸、PS/AE 后期、抠图合成）→ 图生视频/文生视频（首尾帧、运镜、时长、一致性、对口型）→ 商业交付与迭代复现（参数留存、工作流文件复用、版权与可商用判断）」的完整创作-工程-交付决策链。覆盖 (a) 第一性张力 — **可控性/工程化复现（ControlNet/LoRA/ComfyUI 节点工作流/固定种子与参数/可复现可迭代）⇄ 随机性/抽卡玄学（出图随机、prompt 玄学、抽卡刷图、靠运气）**；**审美/创意/艺术指导主导（人的审美判断、概念设计、构图、镜头语言、艺术总监能力——'模型是地板，审美是天花板'）⇄ 工具/模型/参数主导（堆模型能力、堆参数、堆 LoRA、以为换个大模型就行）**；**开源生态/本地可控（Stable Diffusion / ComfyUI / Flux / civitai / liblib 本地部署可定制可控可商用、显卡门槛）⇄ 闭源商业/易用黑盒（Midjourney / 可灵 / 即梦 / Runway 开箱即用但不可精控、按量付费、黑盒）**；**图像生成（T2I/I2I：Midjourney / SDXL / Flux）⇄ 视频生成（T2V/I2V：可灵 / Runway Gen-3 / Vidu / 即梦 / Sora，2024-2025 前沿从图迁移到视频）**；**生产力/商业量产（替代或增强传统设计/插画/电商/广告/影视分镜流程、降本提效、批量）⇄ 纯艺术创作/作者表达（个人风格、实验性、作品集）**；**效率/批量（一次出几十张筛选、商单走量、自动化 pipeline）⇄ 质量/精控（单张精修、商业可用级细节、修手修脸修穿帮）**。(b) 核心工作流 / pipeline（最标准、最易蒸高质量 + CLI 化）— 单图创作链：概念/参考收集（moodboard/风格定位/竞品）→ 模型工具选型（按可控性/风格/成本/可商用判断 MJ vs SD/Flux vs 国产）→ Prompt 与参数设计（主体/风格/光影/镜头/负向词/CFG/采样器/分辨率/种子）→ 首轮出图与方向筛选 → 可控迭代（ControlNet 控构图/姿态/线稿、LoRA 控风格/角色一致性、图生图重绘、inpainting 局部修）→ 放大超分与精修（高清放大/修手修脸/后期调色合成）→ 交付与参数留存（工作流文件/种子/参数复现）。视频创作链：分镜/脚本→首帧出图（图像链产出关键帧）→图生视频（运镜/时长/首尾帧/动态强度）→多镜头一致性（角色/场景/风格统一）→对口型/配音/配乐→剪辑合成→交付。ComfyUI 工程链：节点工作流搭建（加载器/采样/ControlNet/放大/面部修复节点）→ 参数化与复用→批量队列→工作流分享（civitai/openart workflow json）。(c) 工具栈 — 闭源图像（Midjourney v6/v7、DALL·E 3、Adobe Firefly、Ideogram、Recraft）、开源图像与底座（Stable Diffusion 1.5/SDXL/SD3、Flux.1 dev/schnell（Black Forest Labs）、ComfyUI（节点式）、Automatic1111/Forge WebUI、Fooocus）、可控生成插件（ControlNet、LoRA、IPAdapter、AnimateDiff、inpainting/outpainting、ReActor 换脸、面部修复 CodeFormer/GFPGAN、放大 ESRGAN/SUPIR/Topaz）、模型/资源社区（Hugging Face、Civitai、liblib 哩布、吐司 Tusi、openart）、视频生成（可灵 Kling、即梦 Dreamina/Jimeng、Runway Gen-3/Gen-4、Luma Dream Machine、Pika、Vidu、海螺 Hailuo、Sora、Wan 通义万相、MiniMax）、辅助（Magnific 放大、Krea、Photoshop/After Effects 后期、ChatGPT/Claude 写 prompt、reverse prompt 反推）。(d) 知识正典 — AIGC 创作 canon 横跨论文/工程/社区教程：奠基论文（DDPM、Latent Diffusion/Stable Diffusion（Rombach et al）、DiT、ControlNet（Zhang et al）、LoRA、DreamBooth、Textual Inversion、SDEdit、IP-Adapter、视频扩散 SVD/Sora 技术报告）、官方文档与模型卡（Stability AI、Black Forest Labs Flux、ComfyUI 文档、Midjourney docs、可灵/即梦官方教程、Hugging Face diffusers）、社区教程与工作流（ComfyUI 官方示例、civitai 文章、B站/YouTube 创作者长教程、openart workflow）、prompt 工程资源（Midjourney 风格库、提示词指南）。(e) figures/流派 — 模型/工具创造者（Midjourney David Holz、Stability AI Emad/Robin Rombach、Black Forest Labs（Flux，原 SD 团队）、ComfyUI comfyanonymous、ControlNet 张吕敏 Lvmin Zhang、AUTOMATIC1111、可灵/即梦/快手字节团队）、创作者 KOL 与教育者（Midjourney/SD 头部创作者、ComfyUI 工作流大神、AI 视频创作者、国内 B站/抖音 AIGC 教程作者、独立 AI 艺术家）、行业分析与评测（AI 工具评测、生成式 AI 创作生态观察者）。流派分歧：开源可控派（SD/ComfyUI/Flux 本地精控）vs 闭源易用派（MJ/可灵开箱即用）、工程化派（节点工作流/可复现/参数化）vs 玄学抽卡派（prompt 玄学/堆词/抽卡）、审美主导派（'模型是地板审美是天花板'）vs 工具主导派（堆模型堆参数）、图像派 vs 视频派、艺术创作派 vs 商业生产力派、国产工具派（可灵/即梦/liblib）vs 海外工具派（MJ/Flux/Runway）。(f) 行业话术/黑话 — 出图 / 抽卡 / 刷图 / 炼丹（训练 LoRA）/ 喂图 / 垫图（参考图/图生图）/ 控图 / 锁种子 seed / CFG / 采样器 sampler / 步数 steps / 重绘幅度 denoise / 提示词 prompt / 咒语 / 负向 negative / 权重 / tag / 大模型 checkpoint / 底模 / LoRA / ControlNet / 控线稿/openpose/depth / IPAdapter / inpainting 局部重绘 / outpainting 扩图 / 高清放大 / 超分 / 修手 / 崩手 / AI 味 / 风格 / 角色一致性 / 工作流 workflow / 节点 / 跑图 / 队列 / 文生图 T2I / 图生图 I2I / 文生视频 T2V / 图生视频 I2V / 首帧/尾帧 / 运镜 / 动态强度 / 时长 / 对口型 / 一致性 / 可商用 / 训练集 / 过拟合 / 翻车 / 穿帮。(g) 争议/批判 — 「版权与训练数据」（模型训练用未授权作品、生成图版权归属、可商用风险、洗图/伪原创）、「AI 替代创作者就业」（替代画师/插画师/设计师/摄影、行业冲击与转型）、「AI 味与审美趋同」（生成图同质化、套模板、缺乏原创审美）、「玄学 prompt 神话与割韭菜」（'咒语'神秘化、'7天 AIGC 变现/月入过万'付费课割韭菜、抽卡当能力）、「工具快速迭代与淘汰焦虑」（模型月月换、工作流频繁失效、追新疲劳）、「开源 vs 闭源之争」（可控可商用 vs 易用黑盒）、「质量天花板」（修手修脸穿帮、长视频一致性、可控性仍有限）。(h) 大量隐性创作手艺 + 审美 + 工程 + 沟通软技能（prompt 的描述功底与审美词汇、参数与采样的手感、ControlNet/LoRA 的组合控制经验、节点工作流的搭建与调试、批量筛选的审美眼力、修图合成的后期功底、镜头语言与运镜设计、角色/风格一致性的控制、商业需求的翻译与交付、版权与可商用的判断）是高 tacit、靠大量实操跑图 + 审美积累 + 工程调试经验的核心。水分极高（大量'咒语速成/一键变现/AI 绘画割韭菜'营销
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use when adding a new model or pipeline to diffusers, setting up file structure for a new model, converting a pipeline to modular format, or converting weights for a new version of an already-supported model.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use before opening a PR, or whenever asked to self-review a diffusers contribution. Applies the same rubric as the `@claude` CI (checks the diff against references/review-rules.md, traces call paths for dead code). Reports findings grouped by severity, flagging what to fix before submitting (blocking issues + dead code) vs what to leave for the actual review. Report-only — does not edit files.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Implement a generative diffusion model (DDPM or score-based) with noise scheduling, U-Net architecture, training loop, and sampling procedures including DDIM acceleration. Use when building a generative model for image, audio, or molecular synthesis; implementing DDPM from a research paper; adding a custom noise schedule or conditioning mechanism; replacing a GAN-based generator with a diffusion alternative; or prototyping before scaling with production frameworks like diffusers.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
在实现适配器前对候选模型做分析。确定模型实现来源（transformers / 模型目录 / diffusers）、结构特征、内存约束下的逐层量化建议（可选）及 MoE 融合权重风险。适用于用户询问模型适配可行性或做适配前分析时使用。识别为 DiT 扩散模型时，索取推理仓路径后回传 `next_step: model-adapt`，由 orchestrator 委派 `msmodelslim-model-adapt`（DiT 扩展节承接）。
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.