1580158
/
stable-diffusion-webui
mirror of https://github.com/AUTOMATIC1111/stable-diffusion-webui
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
stable-diffusion-webui/docs/milestones/M06/M06_plan.md

5.1 KiB
Raw Blame History

M06 Plan — Prompt / Seed Preparation Extraction

Project: Serena
Phase: Phase II — Runtime Seam Preparation
Milestone: M06
Title: Prompt / seed prep extraction
Posture: Behavior-preserving refactor
Target: Extract prompt + seed preparation logic from processing.py

1. Intent / Target

modules/processing.py currently performs multiple responsibilities inside process_images() and process_images_inner() including:

  • prompt parsing
  • seed preparation
  • subseed logic
  • batch seed generation
  • variation seed handling
  • negative prompt expansion

These steps are mixed with sampling and decoding logic.

The goal of M06 is to extract prompt and seed preparation into a dedicated module while preserving the exact runtime behavior.

This milestone does NOT change generation semantics.

It only relocates preparation logic behind a clean function boundary.

This prepares for:

  • M07 — opts snapshot introduction
  • M08 — process_images_inner snapshot threading
  • M09 — execution context/state seam

2. Scope Boundaries

In Scope

Extract logic related to:

  • prompt parsing
  • negative prompt preparation
  • seed normalization
  • subseed handling
  • seed list generation
  • batch seed assignment
  • seed resizing logic
  • variation strength seed mixing

Target location:

modules/prompt_seed_prep.py

Expose a pure helper function:

prepare_prompt_seed_state(p: StableDiffusionProcessing)

Return a small structured result object.

Out of Scope

The following must remain unchanged:

  • sampler execution
  • conditioning generation
  • latent creation
  • decode pipeline
  • model/VAE loading
  • token merging logic
  • override_settings seam (M05)
  • extension hooks

No API/UI behavior changes are allowed.

3. Invariants

The following behaviors must remain identical.

Prompt handling

  • prompt lists identical
  • negative prompt behavior unchanged
  • prompt matrix logic preserved
  • prompt parser behavior unchanged

Seed handling

  • seed generation identical
  • subseed behavior identical
  • variation seeds identical
  • batch seeds identical
  • deterministic runs unchanged

Extension compatibility

Extensions reading prompt or seed fields must behave identically.

API compatibility

txt2img / img2img endpoints must return identical responses.

Generation determinism

Given identical inputs, outputs must remain identical.

4. Verification Plan

CI Verification

Existing CI gates remain unchanged.

Required passing checks:

Smoke Tests
Linter
Quality Tests
Coverage ≥ 40%

Runtime verification

Smoke tests already exercise:

txt2img
img2img

which invoke the full generation pipeline.

If prompt or seed handling breaks, smoke tests will fail.

Local verification (recommended)

Cursor should run:

pytest test/smoke
pytest test/quality

5. Implementation Steps

Step 1 — Identify extraction block

Locate code inside modules/processing.py responsible for:

  • prompt normalization
  • negative prompt handling
  • seed initialization
  • subseed preparation
  • batch seed generation

These sections should be grouped logically.

Step 2 — Create new module

Create:

modules/prompt_seed_prep.py

Add:

prepare_prompt_seed_state(p)

This function should:

  1. Read prompt + seed values from p
  2. Perform all current preparation logic
  3. Return a structured result

Example structure:

PromptSeedState
prompts
negative_prompts
seeds
subseeds
subseed_strength
seed_resize

The structure should mirror the values currently produced in processing.py.

Step 3 — Replace inline logic

Replace the extracted code in processing.py with:

state = prepare_prompt_seed_state(p)

Then reference fields from state.

Step 4 — Maintain identical behavior

Important:

Do NOT change:

  • variable names used by extensions
  • field assignments on p
  • order of operations
  • default values
  • seed calculation logic

This must be a mechanical extraction only.

Step 5 — Minimal tests

No new tests are required for M06 unless extraction introduces new seams.

Existing smoke + quality tests should cover behavior.

If helpful, a small unit test can be added:

test_prompt_seed_prep.py

but this is optional.

6. Risk & Rollback Plan

Risk: Low (mechanical extraction).

Potential failure modes:

  • prompt list mismatch
  • seed list mismatch
  • extension accessing moved variables

Mitigation:

  • Keep assignments on p
  • Only move internal computation logic

Rollback:

Revert the commit.

Because the change is localized to:

modules/processing.py
modules/prompt_seed_prep.py

7. Deliverables

Expected files changed:

modules/prompt_seed_prep.py      (new)
modules/processing.py            (modified)
docs/milestones/M06/*            (plan, toolcalls, run reports)

No other modules should change.

8. Expected Outcome

After M06:

  • prompt and seed preparation isolated
  • processing.py becomes smaller
  • preparation stage clearly separated from sampling stage

This unlocks:

M07 — Options snapshot introduction M08 — Snapshot threading into process_images_inner M09 — Execution context seam