Lvl 3: Environment / Camera / Setting Change¶
A bigger variant than Lvl 1-2 — this is where you change how the scenes look, not just what's said in them. New environment, different camera angle, different lighting setup, fresh setting.
Unlike Lvl 1-2 (which auto-approves), Lvl 3 goes through a testing phase before it graduates.
When you'd do this¶
- Move the workflow from a "kitchen counter" setting to a "bathroom vanity" setting
- Switch from selfie-style framing to a tripod-mounted medium shot
- Try the same script with the avatar in a car instead of at home
- Add stronger lighting (ring light) where the original used natural window light
- Refresh a winning workflow for a new season (snowy outdoor scene instead of cozy indoor)
If the structure of the workflow changes (add scenes, remove scenes, new prehook) — that's Lvl 4.
What you do¶
You: Lvl 3 variant of XYZS1-V1 — move the setting from kitchen counter
to bathroom vanity. Keep the dialogue, keep the scene count, keep
the avatar wardrobe. Just change where they are.
Claude:
Lvl 3 variant scoped. Confirming direction:
Original setting: kitchen counter, natural window light from
camera-left, minimalist countertop
New setting: bathroom vanity mirror, ring light + ambient bathroom
light, tiled wall behind, vanity counter
Dialogue: unchanged
Wardrobe: unchanged
Scene count: unchanged
Lvl 3 variants enter a testing phase under the next major's -0-N
suffix. Starting at V2-0-1 (testing toward V2).
Image Prompter and Veo Prompter are rewriting the prompts to use
the new setting. Storyboard is being updated to match.
Proceed?
Once you confirm, the work happens:
- Visual Planner updates the storyboard with the new setting
- Image Prompter rewrites the per-scene image prompts
- Veo Prompter updates the per-scene video prompts where the setting affects motion
- PatchWork Importer (Patch Mode) updates the existing
.nbflowwith the new prompts
The output lands at XYZS1-V2-0-1.nbflow in testing/. Not in approved/. Not yet.
What changes in the .nbflow¶
A lot more than Lvl 1-2. Touched:
- Image prompts — almost every scene's image prompt is rewritten (new environment, lighting, camera)
- Video prompts — may be updated (some motion description depends on setting)
- Setting / background descriptions in the storyboard
- Lighting descriptions in image prompts
- Camera angle / framing notes if camera style changed
Unchanged: - Dialogue text - Scene count and structure - Avatar reference sheets - Per-account wardrobe (unless wardrobe is part of the change)
Why a testing phase¶
Lvl 1-2 changes are predictable — dialogue swaps and wardrobe edits land cleanly almost every time. Lvl 3 changes are unpredictable. A new setting means:
- New compositions you haven't seen before
- New lighting that might interact weirdly with the avatar's face
- New camera angles that might not flatter the subject
- New props that might introduce unexpected AI tells
You need to see the output before approving. Hence: testing phase.
Version naming during testing¶
Lvl 3 variants get a V{next_major}-0-N suffix during testing:
V1 → V2-0-1 Lvl 3 variant of V1 enters testing
V2-0-1 → V2-0-2 iteration during testing
V2-0-2 → V2-0-3 another iteration
...
V2-0-N → V2 graduates to V2 in approved/ when you approve
This mirrors how new workflows go through V0-N → V1. Lvl 3 variants are big enough to deserve the same testing rigor.
Workflow during testing¶
After the variant is built:
- Run the Generation Runner on
V2-0-1.nbflow - Review in PatchWork — look at the new setting in action
- Identify scenes that need work
- Either:
- Edit prompts and bump to
V2-0-2 - Use prompt-tuning on stubborn scenes
- Wait for the auto-rerun to handle AI tells
- Edit prompts and bump to
- Iterate until the test account looks clean
- Approve → graduate to
V2inapproved/
You can stay in testing for 5+ iterations if needed. There's no upper bound.
Iteration example¶
V2-0-1: first build with new setting. 2 scenes look great. 3 need work.
Avatar's face is too darkened by the new lighting in Scene 03,
Scene 05's composition feels cramped, Scene 07's background
is busier than the workflow's aesthetic.
V2-0-2: editing image prompts:
Scene 03 — add "soft fill light from camera-front to lift the
face out of shadow"
Scene 05 — wider framing, pull back camera
Scene 07 — simpler background description (matte wall instead
of tiled)
Regen the 3 scenes. Other 5 stay cached.
V2-0-3: 5/8 scenes good. Scene 03 still slightly dim. Run prompt-tuning
on Scene 03 specifically (4 iterations).
V2: all 8 scenes clean. User approves. Graduate to V2 in approved/.
Each iteration bumps the V2-0-N suffix and archives the previous version to backups/. Standard.
When to graduate¶
A Lvl 3 variant graduates to V{N+1} (drop the -0-N suffix) when:
- The test account's outputs are clean (3 of 4 acceptable candidates per scene)
- No AI tells the user is willing to accept
- The user explicitly approves the variant
Approval is required. The pipeline doesn't auto-graduate.
After graduation, the variant is ready for fan-out (same flow as a new workflow — see Chapter 6 — Fan-out).
Lvl 3 across multiple accounts¶
A common pattern: you do the Lvl 3 work on Account A, get it to graduate to V2, then fan out V2 to the other accounts.
V1 (all accounts approved) →
V2-0-1 (Account A testing) →
V2-0-2 (Account A iteration) →
V2-0-3 (Account A iteration) →
V2 (Account A approved) →
V2-1 ? (or fan-out to other accounts in V2 directly)
The fan-out itself can stay as part of V2 (single file, multiple tabs added) or bump to V2-1 if the fan-out introduces meaningful per-account changes. Use judgment.
When you're ready¶
→ Next: Lvl 4: Structural Change. Even bigger — adding / removing / splitting scenes, adding a new prehook, reshaping the workflow's bones. Same testing-phase logic applies.