Model Registry
artokun/comfyui-mcp
Curated download URLs and target directories, organized by family (Flux, WAN, LTX, Qwen, Z-Image, SD15/SDXL), for every model the comfyui-mcp skills reference, covering checkpoints, VAEs, text…
Plan or review LoRA and edit-training work specifically for FLUX.2 Klein or Qwen-Image-Edit, including paired datasets, trainer-version contracts, and held-out fidelity checks.
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-training --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .claude/skills/flux2-lora-training && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .claude/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-trainingType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-training --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .agents/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .agents/skills/flux2-lora-training && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .agents/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-training --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .cursor/skills/flux2-lora-training && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .cursor/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/AnastasiyaW/codex-claude-code-config.git --path skills/ai-ml/flux2-lora-training--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-training --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .gemini/skills/flux2-lora-training && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .gemini/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-trainingInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .github/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .github/skills/flux2-lora-training && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .github/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install AnastasiyaW/codex-claude-code-config flux2-lora-training --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/AnastasiyaW/codex-claude-code-config.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skills/ai-ml/flux2-lora-training .opencode/skills/flux2-lora-training && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "flux2-lora-training" agent skill from https://github.com/AnastasiyaW/codex-claude-code-config/tree/main/skills/ai-ml/flux2-lora-training into .opencode/skills/flux2-lora-training/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "flux2-lora-training", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
flux2-lora-trainingPlan or review LoRA and edit-training work specifically for FLUX.2 Klein or Qwen-Image-Edit, including paired datasets, trainer-version contracts, and held-out fidelity checks.
Flux2 Lora Training is an agent skill from AnastasiyaW/codex-claude-code-config. Plan or review LoRA and edit-training work specifically for FLUX.2 Klein or Qwen-Image-Edit, including paired datasets, trainer-version contracts, and held-out fidelity checks. Do not use for generic Stable Diffusion/DiT training, prompt authoring, or model serving; route those tasks to their specialized skill.
Its SKILL.md is about 4.5k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.
It sits in AI & LLM Engineering, covering Diffusion and image models, Fine-tuning and LLM inference and serving. It works with Qwen and Stable Diffusion. The repository describes itself as: Claude Code, Codex, and multi-agent configuration system: principles, hooks, skills, and workflow patterns for AI-assisted development. The licence is MIT.
3 steps, taken from the first numbered list in SKILL.md.
Read from SKILL.md and the folder at commit 67709af. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Shell commands in SKILL.md call:
pythonFrom the folder's file list and the shell code blocks in SKILL.md.
Links to these hosts (documentation or services it may open):
github.comhuggingface.cohf.coarxiv.orgdocs.bfl.mlmedium.comFrom URLs in SKILL.md, links to its own repository left out.
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Flux2 Lora Training loads about 4.5k tokens when it runs. Until then it costs about 83 tokens; SKILL.md has 1,704 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
The full file from AnastasiyaW/codex-claude-code-config at commit 67709af, republished under its MIT licence (© AnastasiyaW). 1,704 words, ~4,516 tokens.
.claude/skills/flux2-lora-training/SKILL.md (or your agent's skills folder).This file is a route to evaluate a training plan, not a tested recipe. Model cards establish model identity and license; trainer repositories establish only the behavior of their pinned version. Numeric settings and BFS notes below are historical local experiment candidates unless an exact source is linked. Before training, record the model revision, trainer commit, dataset contract, hardware, and a small held-out acceptance set. Do not promote a historical score, memory estimate, slot order, or private result into a universal default.
| Параметр | Klein 9B | Klein 4B | Qwen-Image-Edit | FLUX.1 dev |
|---|---|---|---|---|
| Blocks | 32 (8+24) | 25 (5+20) | 60 (MM-DiT) | 56 (8+48) |
| Embedding dim | 12,288 | 7,680 | — | 15,360 |
| VAE latent channels | 32 (128 packed features) | 32 (128 packed features) | verify the exact revision | 16 |
| Text encoder | Qwen3 (bundled) | Qwen3 | Qwen2.5-VL (7B) | Mistral-Small-3.1 |
| Guidance embeddings | НЕТ | НЕТ | — | Есть |
| Total params | 9B | 4B | 20B DiT + 7B VL | 12B |
FLUX.2 Klein VAE: AutoencoderKLFlux2, 32 latent channels (FLUX.1: 16) → после 2×2 patch packing в трансформере: 32×4 = 128 dims per token, 16× spatial compression. Несовместим с FLUX.1 LoRA — другой VAE, другой latent space. Tiling для больших разрешений: 1024px тайлы с 25% overlap, обрабатывает произвольное разрешение.
Klein editing механизм (Kontext-style): reference image VAE-кодируется и конкатенируется с noise latent вдоль sequence dim. Positional embeddings разделяют reference и output через 3D RoPE time offsets (ref1=t:1, ref2=t:2, output=t:0). Поддерживает до 10 reference images теоретически, обучен на 2.
Text encoder Klein: Qwen3 (встроен в 9B), выходы из слоёв 9, 18, 27.
Do not assume a trainer’s flux_guidance_* fields are no-ops or active from
their names. Verify their behavior against the pinned trainer release and the
selected model before relying on them.
Для LoRA тренировки: base модель FLUX.2-klein-base-9B, не distilled 4-step.
Архитектурно совершенно другая модель. 20B MMDiT (Multimodal DiT) + Qwen2.5-VL 7B как visual-language encoder.
Dual encoding — ключевое отличие от Klein:
Входное изображение →┬→ Qwen2.5-VL 7B → семантические фичи (кто/что/стиль)
└→ VAE encoder → reconstructive latent (текстуры/пиксели)
↓ оба пути сходятся в MMDiTThis is an architectural interpretation, not a measured cross-model guarantee. Assess identity preservation on the task’s held-out images rather than claiming that one model drifts less at an equal LoRA rank.
Consistency Mode (qe2511_consis_alpha) — результат I2I reconstruction training objective: модель дообучена реконструировать входное изображение без изменений. Это выравнивает VL и DiT latent spaces, делая консервативные правки более точными. Триггер restore image details активирует этот bias.
LoRA target modules для Qwen-Image-Edit-2511 (13 модулей, в отличие от Klein):
to_q, to_k, to_v, to_out.0
add_q_proj, add_k_proj, add_v_proj, to_add_out
img_mlp.net.2 ← image-specific FFN (в Klein нет разделения)
img_mod.1 ← image AdaLayerNorm modulation
txt_mlp.net.2 ← text-specific FFN
txt_mod.1 ← text modulation
linear_in, linear_outQwen нативно поддерживает 3 control image слота (control_1, control_2, control_3) через sequence concatenation + MSRoPE.
dataset/
control_1/ ← "before" изображение (источник)
control_2/ ← опциональный 2-й референс (напр., лицо донора)
targets/ ← "after" изображение (результат)
captions/ ← .txt файл с инструкцией на сэмплКритично: имена файлов должны совпадать между папками (0001.png ↔ 0001.png ↔ 0001.txt).
Do not infer folder_path/control_path_1 semantics from issue #536 or PR
#629. The primary records do not establish a reversed-path bug: #536 was
closed as not planned, while #629 fixes Flux2 training that ignored a single
control image. Pin the trainer commit and run a small non-production fixture
that proves which input becomes which control/target before a training job.
Sources: https://github.com/ostris/ai-toolkit/issues/536 and
https://github.com/ostris/ai-toolkit/pull/629
Порядок входных изображений важен. Для BFS Head Swap:
[face, body][body, face] — инвертирование порядка дало значительный прирост качестваHistorical BFS observation: the recorded dataset counts are not a minimum or maximum for another edit. Start from a representative, consented dataset and use held-out fidelity/attribute checks to decide whether more diversity or cleaner pairs are needed.
zero_cond_t трактует control images как чистые референсы при t=0, пока основное изображение следует нормальному diffusion schedule. Предотвращает identity drift.
Use it only when the pinned trainer and inference implementation document the field; then keep train/inference conditioning consistent and verify it on a held-out edit. It is not a universal Qwen-Image-Edit switch.
Trigger ОБЯЗАТЕЛЬНО ставить в user message (slot prompt), НЕ в system/instruction message. LoRA-триггеры обучаются в user message контексте. Это объясняет почему в head swap воркфлоу head_swap: идёт в prompt, а не в instruction.
learning_rate: 1e-4 # снизь до 5e-5 при нестабильности
batch_size: 1
gradient_accumulation: 1-4
lora_rank: 16-32 # стандарт; 64-128 для сложных edits
steps: 2000-6000 # зависит от датасета
save_every: 250-500
optimizer: adamw8bit
dtype: bf16
noise_scheduler: flowmatchBFS Head Swap для Klein 9B конкретно:
python train.py \
--learning_rate 1e-4 \
--epochs 5 \
--lora_rank 32 \
--max_pixels 1048576 # 1024×1024LoRA target modules (24 transformer blocks):
to_q, to_k, to_v, to_out.0,
add_q_proj, add_k_proj, add_v_proj, to_add_out,
linear_in, linear_out, to_qkv_mlp_projПолный fine-tune скрипт: examples/flux2/model_training/full/FLUX.2-klein-base-9B.sh
{
"model_family": "flux2",
"model_flavour": "klein-9b",
"base_model_precision": "int8-quanto",
"lora_rank": 16,
"learning_rate": 1e-4,
"lr_scheduler": "constant",
"flux_guidance_mode": "constant",
"flux_guidance_value": 1.0,
"gradient_checkpointing": true
}VRAM: ~14GB (int8-quanto), ~22GB (bf16).
Что реально работает:
linear: 128, linear_alpha: 64, conv: 64, conv_alpha: 32 → улучшает в почти каждом тестеlr: 0.000095, timestep_type: shift, weight_decay: 0.00015Обучаешь модель на парах где маскированные области закрашены чистым зелёным (RGB 0, 255, 0). Модель учит: зелёный = "перегенерировать эту область по тексту".
control_image slot в ai-toolkit — не нужна модификация архитектурыostris/qwen_image_edit_inpainting на HuggingFacemask_conditioning_type — добавляет proper mask-conditioned training для FLUX.2 Klein.
Подходит для более строгого контроля маски.
FLUX.1-Fill approach (concatenated latents, 16+16+64 channels) требует полной модификации архитектуры модели — не реализуемо как LoRA. Нужен full fine-tune с добавлением input channels.
Klein официально поддерживает 64×64 до 4MP (2048×2048), dimensions кратны 16.
--max_pixels| Разрешение | max_pixels |
|---|---|
| 1024×1024 (1MP) | 1048576 |
| 1536×1536 | 2359296 |
| 2048×2048 (4MP) | 4194304 |
Нет архитектурных ограничений — Klein использует RoPE (resolution-agnostic).
| Разрешение | VRAM (bf16) | Оптимизации |
|---|---|---|
| 1024×1024 | ~22GB | базовые |
| 1536×1536 | 40–80GB+ | grad_checkpoint + fp8 + 8bit adam |
| 2048×2048 | 80GB+ | все + cache_latents |
H200 (140GB) комфортно берёт 1536 с оптимизациями.
Начинай с маленького разрешения (256–512px), постепенно увеличивай. Прыжок сразу на 1536 может вызвать нестабильность. Kohya-ss задокументировали деградацию likeness при >1024px без curriculum подхода.
{
"resolution_type": "pixel_area",
"crop_aspect": "square",
"crop_style": "center"
}crop_aspect: square — наиболее надёжный вариант для multi-aspect.
Нет опубликованных методов расширения native resolution Klein через checkpoint fine-tuning (на март 2026).
На практике это не нужно — Klein уже генерирует до 4MP нативно.
Если всё же нужен "latent expansion":
--max_pixels через DiffSynth-Studio full fine-tuneFull checkpoint fine-tune скрипт: examples/flux2/model_training/full/FLUX.2-klein-base-9B.sh (DiffSynth-Studio)
| Фреймворк | Klein 9B | Edit LoRA | Inpaint | Multi-Aspect | Заметки |
|---|---|---|---|---|---|
| ai-toolkit | Да | Да (native) | Green screen | Ограничено | Лучший для edit LoRA |
| SimpleTuner | Да | Частично (PR#2520) | Mask conditioning | Да (с нюансами) | Лучшая документация |
| DiffSynth-Studio | Да | Да (ICEdit format) | Не задокументировано | Да (dynamic) | Поддерживает full fine-tune |
| fal.ai trainer | Да (hosted) | Да (Qwen 2511) | Нет | Нет | Cloud, проще всего |
| kohya-ss | Ограничено | Нет | Нет | Да | Нестабильно для Klein |
Рекомендация для edit LoRA (head swap, face swap): ai-toolkit + base модель.
Банальная тайловая обработка через ComfyUI-TiledDiffusion даёт цветовые скачки между тайлами: каждый тайл генерируется независимо без знания соседей.
Klein нативно поддерживает произвольное число reference images через sequence concatenation + 3D RoPE. Это можно использовать для тайлов:
Tile 1 (обработан) → reference_1
Tile 2 (обработан) → reference_2
Tile 3 (текущий) → output (Klein видит соседей при генерации)Практический порядок обхода (BFS-style):
Это работает за счёт того, что Klein использует те же RoPE time offsets что и для head swap: ref1=t:1, ref2=t:2, output=t:0. Тайлы по-сути становятся "видеокадрами" в пространстве последовательности.
| Метод | Где | Принцип | Для FLUX/DiT |
|---|---|---|---|
| MultiDiffusion | ComfyUI-TiledDiffusion | Усреднение latents в overlapping зонах | Работает (базовый) |
| SyncDiffusion | отдельно | LPIPS loss для согласованности цветов | Нет реализации для Klein |
| SpotDiffusion | ComfyUI-TiledDiffusion | Shifting windows, снижает артефакты | Лучший из TiledDiffusion |
| DemoFusion | отдельно | Progressive upscaling + skip residuals | Нет для Klein |
| DyPE | отдельно | Training-free 16MP для FLUX | Только для FLUX.1 |
| Scale-DiT | arxiv:2501.12900 | LoRA для масштабирования DiT | Обучается |
| Kontext-concat | нативно в Klein | Соседние тайлы как references | Лучший |
Большое изображение (напр. 3072×2048)
↓
VAEEncodeTiled (1024px тайлы, 25% overlap)
↓
Для каждого тайла:
[уже_обработанные_соседи] → LoadImage → reference_1, reference_2
[текущий_тайл] → noise latent
Klein KSampler с 2 references
↓
VAEDecodeTiled
↓
Stitch с Gaussian blending в зонах overlapКлючевые ноды: ComfyUI-TiledDiffusion (SpotDiffusion режим), VAEEncodeTiled, VAEDecodeTiled.
FLUX.1/Klein трансформер не содержит conv слоёв в основной части — только attention и MLP. Circular padding применимо только к VAE (там есть conv), но не к самому DiT. Поэтому бесшовных панорам через circular padding не получить — используй StitchDiffusion LoRA.
| Сценарий | Решение |
|---|---|
| Изображение ≤ 4MP | Klein напрямую (поддерживает до 2048px нативно) |
| Изображение > 4MP | VAEEncodeTiled + SpotDiffusion |
| Панорама (seamless) | StitchDiffusion LoRA + wrap padding в VAE |
| Batch крупных продуктовых кадров | Тайлинг с Kontext-соседями |
Нет. Qwen-Image-Edit-2509/2511 уже нативно поддерживает 2 control image слота. Для head swap (body + face → result) архитектурных изменений не требуется:
dataset/
control_1/ ← BODY (поза, одежда, окружение)
control_2/ ← FACE (лицо донора)
targets/ ← RESULT (голова пересажена)
captions/ ← "head_swap: ..."Порядок слотов критичен и должен быть консистентным во всём датасете.
Только добавление control_3/ папки + новый time offset в position embeddings (t=3). Веса трансформера не меняются. Это уже поддерживается в Qwen-Image-Edit нативно.
Проблема 2-slot LoRA: модель иногда берёт волосы с body вместо face. Решения:
A) Текстовые якоря (текущий подход): явно в промпте "hair MUST match Picture 2", повторить 2-3 раза в разных формулировках.
B) FuseAnyPart-style adapter (arxiv:2410.22771): отдельный lightweight adapter принимает masked_face + masked_body → combined conditioning. Жёсткое маскирование на уровне латентов. Требует нового компонента поверх модели.
C) ACE++ (arxiv:2501.02487): Long-context Condition Unit — все conditioning images (reference + mask + output) конкатенируются в единый extended sequence. Two-stage training: сначала 0-ref, потом multi-ref задачи.
| Аспект | 1-input | 2-input (head swap) |
|---|---|---|
| Датасет | (source, target) пары | (body, face, result) триплеты |
| Синтетика | Можно генерировать | Нужны реальные результаты head swap как GT |
| Caption | Общая инструкция | Явные ссылки на "image 1" и "image 2" |
| Порядок | Не критичен | Строго: body=control_1, face=control_2 |
| zero_cond_t | Желательно | Обязательно (иначе references зашумляются) |
| Проблема | Причина | Решение |
|---|---|---|
| Модель учит трансформацию в обратную сторону | Семантика control/target не подтверждена для закреплённой версии trainer | Зафиксировать commit и проверить малым control/target fixture до training job |
| LyCORIS создаёт 0 модулей | Key mismatch с Klein architecture | Использовать стандартный LoRA format |
| SimpleTuner зависает на RTX 5090 | Issue #2477 | Fallback на H200 |
| Деградация likeness при >1024px | Прямой jump на высокое разрешение | Curriculum: 256→512→1024→1536 |
| identity drift при editing | zero_cond_t не включён | Включить и при тренировке, и при inference |
| Attribute leakage (волосы с body вместо face) | 2-slot LoRA не имеет жёсткого mask control | Текстовые якоря ИЛИ FuseAnyPart adapter |
© AnastasiyaW, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Just SKILL.md in skills/ai-ml/flux2-lora-training of AnastasiyaW/codex-claude-code-config.
Open the folder on GitHubat commit 67709af
Flux2 Lora Training next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Flux2 Lora Training this skillAnastasiyaW/codex-claude-code-config | 154 | — | ~4.5k | Automated safety check: Pass | MIT | |
| Model Registryartokun/comfyui-mcp | 803 | — | ~2.1k | Automated safety check: Pass | MIT | |
| Castguaardvark/guaardvark | 258 | — | ~673 | Automated safety check: Pass | MIT | |
| LoRA Space Builderhuggingface/skills | 11k | 2 repos | ~8.4k | Automated safety check: Pass | Apache-2.0 | |
| Workflow Template BuilderMooshieblob1/MooshieUI | 207 | — | ~640 | Automated safety check: Pass | AGPL-3.0 | |
| Z Image Txt2imgartokun/comfyui-mcp | 803 | — | ~2.8k | Automated safety check: Pass | MIT |
artokun/comfyui-mcp
Curated download URLs and target directories, organized by family (Flux, WAN, LTX, Qwen, Z-Image, SD15/SDXL), for every model the comfyui-mcp skills reference, covering checkpoints, VAEs, text…
guaardvark/guaardvark
Build consistent characters, environments and props in Guaardvark's Cast Library and train LoRAs for them locally (reference photos → vision bible → sample plan → approved samples → training).
huggingface/skills
Builds and publishes a Gradio demo on Hugging Face Spaces for a LoRA, with the pipeline, UI and settings chosen to match that LoRA's task and model card.
Mooshieblob1/MooshieUI
Builds or modifies ComfyUI workflow JSON templates in MooshieUI's Rust backend (src-tauri/src/templates).
artokun/comfyui-mcp
Build Z-Image txt2img workflows. An agent skill from artokun/comfyui-mcp.
artokun/comfyui-mcp
Anime/illustration text-to-image (ANIMA 1.0, ~2B Cosmos DiT).
AnastasiyaW/codex-claude-code-config
Find likely software bugs in a codebase, rank concrete bug candidates, and prove or reject them with focused regression tests before proposing a fix.
AnastasiyaW/codex-claude-code-config
A skill your agent uses when implementing Motion or Framer Motion in React/JavaScript: interactive UI components, micro-interactions, gestures, layout or page transitions, and scroll-based animation.
AnastasiyaW/codex-claude-code-config
Plan-based verification - freeze acceptance criteria before building, then verify after with an independent fresh-context agent (the builder must not verify their own work).
AnastasiyaW/codex-claude-code-config
Написание и запуск Claude Code dynamic workflows (JS-оркестратор субагентов).
AnastasiyaW/codex-claude-code-config
A skill your agent uses when: NotebookLM, notebooklm MCP, large documentation sets, courses, books, papers, or citation-backed research are mentioned.
AnastasiyaW/codex-claude-code-config
Validate a proposed DeepSeek API integration before any key or project context is sent: check thinking-mode tool-call history, strict-schema assumptions, bounded output, and provider data boundaries.
Works with
Categories
Plan or review LoRA and edit-training work specifically for FLUX.2 Klein or Qwen-Image-Edit, including paired datasets, trainer-version contracts, and held-out fidelity checks. Flux2 Lora Training is an agent skill from AnastasiyaW/codex-claude-code-config.2 Klein or Qwen-Image-Edit, including paired datasets, trainer-version contracts, and held-out fidelity checks.
Flux2 Lora Training fits situations like: generic Stable Diffusion/DiT training; prompt authoring; route those tasks to their specialized skill.
Run `npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a claude-code`. Or copy the skill folder (skills/ai-ml/flux2-lora-training in AnastasiyaW/codex-claude-code-config) into .claude/skills/flux2-lora-training in your project. Claude Code loads it when a task matches its description.
Run `npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a codex`. Or copy the skill folder (skills/ai-ml/flux2-lora-training in AnastasiyaW/codex-claude-code-config) into .agents/skills/flux2-lora-training in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add AnastasiyaW/codex-claude-code-config --skill flux2-lora-training -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/flux2-lora-training, .gemini/skills/flux2-lora-training, .github/skills/flux2-lora-training and .opencode/skills/flux2-lora-training in your project.
Going by SKILL.md and its folder, Flux2 Lora Training needs the command-line tools its instructions call (python). Our summary lists: Python 3.
SKILL.md names 6 domains. As links in the text: github.com, huggingface.co, hf.co, arxiv.org, docs.bfl.ml and medium.com. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
Flux2 Lora Training is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
About 4.5k tokens (SKILL.md is roughly 18k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.
Skills that share tags, products or a category with Flux2 Lora Training: Model Registry (artokun/comfyui-mcp, 803 stars), Cast (guaardvark/guaardvark, 258 stars), LoRA Space Builder (huggingface/skills, 11k stars) and Workflow Template Builder (Mooshieblob1/MooshieUI, 207 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
AnastasiyaW (a GitHub user) maintains it in AnastasiyaW/codex-claude-code-config, which has 154 GitHub stars. The repository holds 50 skills in this directory. The repository was last updated on October 9, 2026.
Source: AnastasiyaW/codex-claude-code-config on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.