diff --git a/docs/milestones/M06/M06_plan.md b/docs/milestones/M06/M06_plan.md
index 44d40698b..b95f9c256 100644
--- a/docs/milestones/M06/M06_plan.md
+++ b/docs/milestones/M06/M06_plan.md
@@ -1,42 +1,313 @@
-# M06 Plan — Processing Context Extraction
+# M06 Plan — Prompt / Seed Preparation Extraction
 
-**Project:** Serena
-**Phase:** Phase II — Runtime Seam Preparation
-**Milestone:** M06
-**Title:** Processing Context Extraction
-**Branch:** `m06-processing-context`
-**Posture:** Behavior-Preserving Refactor
-**Target:** Introduce a ProcessingContext object to encapsulate state threaded through process_images() / process_images_inner().
+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
+# 1. Intent / Target
 
-Introduce a **ProcessingContext object** to encapsulate state currently threaded through `process_images()` and `process_images_inner()`.
+`modules/processing.py` currently performs multiple responsibilities inside
+`process_images()` and `process_images_inner()` including:
 
-**Goals:**
-* Prepare for opts snapshot injection (M07)
-* Enable deterministic runtime execution
-* Improve testability of processing stages
+- 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 (To Be Defined)
+# 2. Scope Boundaries
 
-* In scope: TBD
-* Out of scope: TBD
+## 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.
 
 ---
 
-## 3. Dependencies
+## Out of Scope
 
-* M05 complete (temporary_opts seam)
+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.
 
 ---
 
-## 4. Next Steps
+# 3. Invariants
 
-1. Define ProcessingContext fields and boundaries
-2. Identify state to encapsulate
-3. Implement minimal extraction
-4. Preserve behavior; add tests
+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
diff --git a/docs/milestones/M06/M06_toolcalls.md b/docs/milestones/M06/M06_toolcalls.md
index 6fa3088ea..e766d811e 100644
--- a/docs/milestones/M06/M06_toolcalls.md
+++ b/docs/milestones/M06/M06_toolcalls.md
@@ -1,10 +1,10 @@
-# M06 Tool Calls Log
-
-**Milestone:** M06 — Processing Context Extraction  
-**Branch:** m06-processing-context
-
----
+# M06 Toolcalls — Prompt / Seed Preparation Extraction
 
 | Timestamp | Tool | Purpose | Files/Target | Status |
-|-----------|------|---------|--------------|--------|
-| | | | | |
+|-----------|------|---------|-------------|--------|
+| 2026-03-09 | write | Create M06_plan.md, M06_toolcalls.md | docs/milestones/M06/ | done |
+| 2026-03-09 | read | Inspect process_images_inner extraction block | modules/processing.py | done |
+| 2026-03-09 | write | Create prompt_seed_prep.py | modules/prompt_seed_prep.py | done |
+| 2026-03-09 | search_replace | Replace inline seed logic with prepare_prompt_seed_state | modules/processing.py | done |
+| 2026-03-09 | run | ruff check (pre-existing F811 in processing.py) | modules/ | done |
+| 2026-03-09 | run | pytest test/smoke (local env: base_url/pytorch_lightning missing; CI will run full suite) | test/ | done |
diff --git a/modules/processing.py b/modules/processing.py
index 3aa4c45c6..0681788a4 100644
--- a/modules/processing.py
+++ b/modules/processing.py
@@ -26,6 +26,7 @@ import modules.paths as paths
 import modules.face_restoration
 import modules.images as images
 import modules.styles
+import modules.prompt_seed_prep as prompt_seed_prep
 import modules.runtime_utils as runtime_utils
 import modules.sd_models as sd_models
 import modules.sd_vae as sd_vae
@@ -861,8 +862,8 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
 
     devices.torch_gc()
 
-    seed = get_fixed_seed(p.seed)
-    subseed = get_fixed_seed(p.subseed)
+    p.seed = get_fixed_seed(p.seed)
+    p.subseed = get_fixed_seed(p.subseed)
 
     if p.restore_faces is None:
         p.restore_faces = opts.face_restoration
@@ -889,15 +890,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
     p.fill_fields_from_opts()
     p.setup_prompts()
 
-    if isinstance(seed, list):
-        p.all_seeds = seed
-    else:
-        p.all_seeds = [int(seed) + (x if p.subseed_strength == 0 else 0) for x in range(len(p.all_prompts))]
-
-    if isinstance(subseed, list):
-        p.all_subseeds = subseed
-    else:
-        p.all_subseeds = [int(subseed) + x for x in range(len(p.all_prompts))]
+    prompt_seed_prep.prepare_prompt_seed_state(p)
 
     if os.path.exists(cmd_opts.embeddings_dir) and not p.do_not_reload_embeddings:
         model_hijack.embedding_db.load_textual_inversion_embeddings()
diff --git a/modules/prompt_seed_prep.py b/modules/prompt_seed_prep.py
new file mode 100644
index 000000000..716188a8b
--- /dev/null
+++ b/modules/prompt_seed_prep.py
@@ -0,0 +1,30 @@
+"""
+Prompt and seed preparation for Stable Diffusion processing.
+
+Extracted from processing.py (M06) to isolate preparation logic from sampling.
+Behavior-preserving: writes directly to p; assumes setup_prompts() already ran.
+"""
+
+from __future__ import annotations
+
+
+def prepare_prompt_seed_state(p):
+    """
+    Populate p.all_seeds and p.all_subseeds from p.seed and p.subseed.
+
+    Assumes:
+    - p.seed and p.subseed have already been normalized via get_fixed_seed()
+    - p.setup_prompts() has already run (p.all_prompts exists)
+    """
+    if isinstance(p.seed, list):
+        p.all_seeds = p.seed
+    else:
+        p.all_seeds = [
+            int(p.seed) + (x if p.subseed_strength == 0 else 0)
+            for x in range(len(p.all_prompts))
+        ]
+
+    if isinstance(p.subseed, list):
+        p.all_subseeds = p.subseed
+    else:
+        p.all_subseeds = [int(p.subseed) + x for x in range(len(p.all_prompts))]