AGENT SKILL

deep-daze

Use deep-daze for CLIP-guided Siren image generation through the imagine CLI and Python APIs, with runtime inspection and troubleshooting.

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

01 前言 Front Matter

name
deep-daze
license
MIT
metadata
{ "disco-role": "operating" }
description
Use deep-daze for CLIP-guided Siren image generation through the imagine CLI and Python APIs, with runtime inspection and troubleshooting.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

Deep Daze

Use this repo skill when a task involves the deep-daze / deep_daze package: CLIP-guided text-to-image generation, image-conditioned generation, start-image priming, story mode, the imagine command, or programmatic Imagine and DeepDaze usage.

Deep Daze combines OpenAI CLIP guidance with a Siren generator. Generation is an optimization run, not a one-shot inference call: it can download CLIP weights, consume substantial VRAM or CPU time, and write image/progress artifacts in the current working directory.

Quick install and identity checks

Install the public package in the user's environment:

python -m pip install deep-daze

Minimal import and package-data check:

python - <<'PY'
from deep_daze import DeepDaze, Imagine
from deep_daze.clip import available_models, tokenize
print(DeepDaze.__name__, Imagine.__name__)
print(available_models())
print(tuple(tokenize('a house').shape))
PY

Expected CLIP model names are RN50, RN101, RN50x4, ViT-B/32, and ViT-L/14; ViT-B/32 is the default model. tokenize('a house') should return shape (1, 77).

For a safe inspection that does not download model checkpoints or generate images, run the bundled helper:

python scripts/deep_daze_inspect.py

Read references/usage-overview.md for the package concept map and references/troubleshooting.md for cross-cutting failures.

Route by task

User task Read next
Verify install/import, dependency presence, CLIP model names, tokenizer package data, cache/download behavior, CPU/GPU backend status, or VRAM expectations. sub-skills/runtime-and-models/SKILL.md
Build or debug an imagine command for text prompts, image targets, text+image conditioning, start-image priming, story mode, progress frames, GIF/video output, or headless execution. sub-skills/cli-workflows/SKILL.md
Use deep_daze.Imagine or lower-level DeepDaze from Python, translate CLI settings into code, inspect signatures/defaults, use custom CLIP encodings, or control output paths programmatically. sub-skills/python-api/SKILL.md

Safe operating defaults

  • Start with small epochs, iterations, image_width, and batch_size until the prompt, output path, and runtime are known to work.
  • Use open_folder=False in scripts, notebooks, remote shells, and headless agents.
  • Use save_date_time=True or a clean output directory to avoid CLI overwrite prompts.
  • Keep ordinary prompts within CLIP's 77-token context. Use story mode for long prose, with save_progress=True when the transition frames matter.
  • Treat full generation as a potentially long/networked run. It may download a CLIP checkpoint into the user's cache before training starts.

Useful entry points

  • CLI: imagine "a house in the forest" --open_folder=False
  • Python: from deep_daze import Imagine
  • Runtime CLIP utilities: from deep_daze.clip import available_models, tokenize
  • Lower-level generator module: from deep_daze import DeepDaze

Validation guidance

Use the bundled inspection helpers and CLI/API probes before expensive runs:

Provenance and refresh

Read references/repo-provenance.md before using this skill against a different checkout or package version. If the source commit, dirty state, package version, public entry points, CLI signature, or core source layout changed, refresh the repo skill before relying on details.

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