AGENT SKILL

inference

Routes DreamOmni2 command-line image editing and generation workflows.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
inference
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Routes DreamOmni2 command-line image editing and generation workflows.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

DreamOmni2 inference

Use this sub-skill when the user wants to run the DreamOmni2 CLI workflows that take two images plus an instruction and produce either an edited image or a newly generated image.

Typical triggers

  • "generate an image with DreamOmni2"
  • "edit this image with DreamOmni2"
  • "run the generation CLI"
  • "run the editing CLI"
  • "what do the two input images mean in the DreamOmni2 scripts?"

What belongs here

  • The two-image editing workflow
  • The two-image generation workflow
  • The VLM prompt stage that turns the instruction into a diffusion prompt
  • The diffusion pipeline invocation and saved output path
  • CLI validation and prompt/input-order troubleshooting for these workflows

What does not belong here

  • Gradio launchers and browser-facing UI issues; use sub-skills/web-demo/
  • Training, checkpointing, or distributed optimization helpers
  • Generic Diffusers or generic VLM tasks that are not DreamOmni2-specific

Read these files first

  • references/workflows.md for the edit vs generation command shapes and input order
  • references/api-reference.md for the verified pipeline and helper signatures
  • references/troubleshooting.md for GPU, model-path, prompt, and output failures
  • ../../scripts/dreamomni2_common.py for the shared prompt/image helpers used by both wrappers

Bundled scripts

  • scripts/inference_edit.py for the editing CLI
  • scripts/inference_gen.py for the generation CLI

Workflow outline

  1. Confirm whether the task is editing or generation.
  2. Confirm the two input images and the instruction text.
  3. Confirm the model paths, especially the VLM checkpoint and the matching LoRA directory.
  4. For editing, make sure the source image is first and the reference image is second.
  5. Run the bundled CLI wrapper and save the output image.
  6. If the result is wrong, inspect the prompt stage and the image order before changing the model stack.

Decision points

  • Editing vs generation: the same VLM prompt stage is used, but editing and generation load different LoRA adapters and have different output expectations.
  • Default size: generation exposes height and width; editing uses the resized source image dimensions.
  • Prompt text: the wrapped VLM stage should preserve the task-specific prefix so the downstream diffusion prompt is well formed.

If you need more detail

Read the linked reference files when you need exact parameter names, supported image types, or a recovery path for a missing model or GPU.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351