Cast.md
... ...
@@ -1,44 +1,60 @@
1 1
# Cast
2 2
3
-**Status:** Candidate visual locks
3
+**Status:** Candidate visual locks revised after `EXP-TLC-001`
4 4
5
-## Lumi
5
+## Mia
6 6
7
-| Field | Lock |
7
+Mia replaces the earlier working character name **Lumi**.
8
+
9
+| Field | Candidate lock |
8 10
| --- | --- |
9 11
| Role | Calm, observant guide; notices what others miss |
10
-| Silhouette | Rounded forms; two clearly separated hair buns |
11
-| Hair | Deep navy / blue-black, twin buns |
12
-| Eyes | Large teal eyes |
13
-| Palette | Teal, cream, warm gold |
14
-| Signature object | Small lantern pendant |
15
-| Movement | Measured, graceful, reassuring |
16
-| Recognition test | Readable from silhouette, close-up, and back three-quarter view |
12
+| Face | Softly rounded, warm brown eyes, light freckles |
13
+| Hair | Straight chestnut-brown chin-length bob, simple side part, one small teal hair clip |
14
+| Wardrobe | Teal zip jacket, cream T-shirt, denim pants, cream sneakers |
15
+| Palette | Teal, cream, denim blue, chestnut brown |
16
+| Signature object | Tiny round lantern pendant; secondary continuity detail until a reference pack proves it stable |
17
+| Movement | Measured, natural, reassuring |
18
+| Recognition test | Readable from hairstyle, teal clip, palette, close-up, and back three-quarter view |
19
+
20
+## Henry
17 21
18
-## Pip
22
+Henry replaces the earlier working character name **Pip**.
19 23
20
-| Field | Lock |
24
+| Field | Candidate lock |
21 25
| --- | --- |
22 26
| Role | Energetic experimenter; acts, reacts, and makes discoveries physical |
23
-| Silhouette | Angular, lively; distinct spiky cowlick |
24
-| Hair | Auburn / chestnut |
25
-| Eyes | Warm amber |
26
-| Palette | Mustard, blue, burnt orange |
27
-| Signature objects | Small satchel and scarf |
27
+| Face | Natural child proportions, warm brown eyes, light freckles |
28
+| Hair | Compact tousled medium-brown hair with one small cowlick |
29
+| Wardrobe | Mustard zip jacket, navy T-shirt, denim pants, simple sneakers |
30
+| Palette | Mustard, navy, denim blue, burnt orange |
31
+| Signature objects | Short burnt-orange neckerchief and small blue cross-body satchel; secondary continuity details until the reference pack proves them stable |
28 32
| Movement | Quick, expressive, slightly impulsive |
29
-| Recognition test | Readable from silhouette, close-up, and back three-quarter view |
33
+| Recognition test | Readable from hair shape, palette, close-up, and back three-quarter view |
30 34
31 35
## Pair contrast
32 36
33
-Lumi is round/cool/calm; Pip is angular/warm/kinetic. Their height, hair shape,
34
-palette, signature object, and movement language must remain separable even when
35
-the model introduces small generational variance.
37
+Mia is calm and teal-led; Henry is kinetic and mustard-led. Their hair shapes,
38
+palettes, wardrobe blocks, and movement languages should remain separable even
39
+when the model introduces small generational variance. Do not prompt an explicit
40
+height relationship; let the shot composition determine their apparent scale.
36 41
37 42
## Character pack required before lock
38 43
39 44
Front, profile, back, three-quarter, neutral close-up, happy close-up, concerned
40
-close-up, full-body action pose, seated pose, Lumi/Pip height comparison, and a
41
-shared neutral two-shot.
45
+close-up, full-body action pose, seated pose, and a shared neutral two-shot.
46
+
47
+Do not use personality adjectives alone in image prompts. Compile only the
48
+visible features needed by the shot. The selected reference images, rather than
49
+the character names, must carry appearance.
50
+
51
+## Current candidate references
52
+
53
+- Mia solo: `Mia-ReadingNook-v007_00001_.png`
54
+- Henry solo: `Henry-Soccer-v007_00001_.png` and
55
+ `Henry-Pond-v007_00001_.png`
56
+- Pair: `Mia-Henry-AutumnWalk-v007_00001_.png` and
57
+ `Mia-Henry-Farm-v008_00001_.png`
42 58
43
-Do not use personality adjectives alone in image prompts. Compile the visible
44
-features needed by the shot.
59
+These are review candidates, not locked character sheets. See
60
+[Character, Style, and I2V Tests — 2026-07-27](/Character-Style-and-I2V-Tests-2026-07-27).
Character-Style-and-I2V-Tests-2026-07-27.md
... ...
@@ -0,0 +1,196 @@
1
+# Character, Style, and I2V Tests — 2026-07-27
2
+
3
+**Experiment ID:** `EXP-TLC-001`
4
+**Status:** Completed exploratory pass; character sheets and style remain
5
+candidate
6
+**Models:** Anima Base v1.0 stills; Wan 2.2 14B I2V with LightX2V four-step
7
+LoRAs
8
+**Review system:** SmartGallery / Media Depot
9
+
10
+## Questions
11
+
12
+1. Can two recurring children remain recognizable across unrelated
13
+ environments using text prompts alone?
14
+2. Which visible anchors survive independent Anima generations?
15
+3. Which prompt patterns produce clean, full-bleed editorial frames?
16
+4. Can the visual family of the Mia Reading Nook image be repeated through a
17
+ fixed text style block?
18
+5. Can the local dual-A4000 ComfyUI system execute two Wan 2.2 14B Turbo I2V
19
+ tests in parallel?
20
+
21
+## Still-image configuration
22
+
23
+| Field | Value |
24
+| --- | --- |
25
+| Runtime | Local dual-instance ComfyUI |
26
+| Model | `anima-base-v1.0.safetensors` |
27
+| Text encoder | `qwen_3_06b_base.safetensors` |
28
+| VAE | `qwen_image_vae.safetensors` |
29
+| Dimensions | 864 × 1152, 3:4 portrait |
30
+| Steps | 30 |
31
+| CFG | 4 |
32
+| Sampler / scheduler | `er_sde` / `simple` |
33
+| Batch strategy | Independent seed per frame; one queue per GPU |
34
+
35
+## Character direction tested
36
+
37
+The first pass used the working names Lumi and Pip and deliberately exaggerated
38
+their contrast: twin navy buns and teal eyes for Lumi; a large auburn cowlick,
39
+scarf, and satchel for Pip. The silhouettes read clearly, but the designs felt
40
+more like stylized mascots than ordinary recurring children. Lumi also drifted
41
+more than Pip across facial structure, skin tone, clothing, pendant size, and
42
+body proportions.
43
+
44
+The owner approved the names **Mia and Henry** and requested familiar,
45
+everyday-child styling.
46
+
47
+### Mia candidate
48
+
49
+- Chestnut-brown chin-length bob with a simple side part.
50
+- One small teal hair clip.
51
+- Warm brown eyes and light freckles.
52
+- Teal zip jacket, cream T-shirt, denim pants, cream sneakers.
53
+- Tiny round pendant as a secondary detail.
54
+- Calm, observant movement language.
55
+
56
+### Henry candidate
57
+
58
+- Compact tousled medium-brown hair with one small cowlick.
59
+- Warm brown eyes and light freckles.
60
+- Mustard zip jacket, navy T-shirt, denim pants, simple sneakers.
61
+- Short burnt-orange neckerchief and small blue satchel as secondary details.
62
+- Energetic, physical movement language.
63
+
64
+Do not prompt a height relationship between Mia and Henry. The experiment
65
+showed that explicit height language can exaggerate scale differences and make
66
+one child read as substantially older.
67
+
68
+## Candidate reference images
69
+
70
+| Purpose | Candidate |
71
+| --- | --- |
72
+| Mia solo | `Mia-ReadingNook-v007_00001_.png` |
73
+| Henry solo | `Henry-Soccer-v007_00001_.png` |
74
+| Henry alternate | `Henry-Pond-v007_00001_.png` |
75
+| Pair | `Mia-Henry-AutumnWalk-v007_00001_.png` |
76
+| Pair alternate | `Mia-Henry-Farm-v008_00001_.png` |
77
+
78
+These are not locked reference sheets. They are the strongest candidates from
79
+this pass for the next reference-guided experiment.
80
+
81
+## Anima findings
82
+
83
+### What worked
84
+
85
+- Strong, simple anchors survive better than personality adjectives.
86
+- Mia's bob, teal clip, teal jacket, brown eyes, and freckles formed a much
87
+ more stable everyday design than the twin-bun concept.
88
+- Henry's brown cowlick, mustard/navy block, and orange accent remained broadly
89
+ recognizable.
90
+- The teal-led Mia and mustard-led Henry remained separable in shared scenes.
91
+- Ordinary, lived-in locations produced stronger emotional readability than
92
+ abstract creation imagery.
93
+- Two GPUs processed independent still candidates reliably in parallel.
94
+
95
+### Failure patterns
96
+
97
+- Solid-color borders, white margins, circular vignettes, floating dioramas,
98
+ and poster-like negative space.
99
+- Two-panel or duplicated scenes when a prompt implied multiple sequential
100
+ actions.
101
+- Clothing, pendant, scarf, satchel, skin tone, eye shape, and body-proportion
102
+ drift under text-only generation.
103
+- Complex hand and prop actions increased anatomy defects.
104
+- Text occasionally appeared in maps, exhibit labels, or artwork despite a
105
+ no-text instruction.
106
+
107
+### Prompt rules adopted
108
+
109
+1. Begin with `single continuous camera image, one frame, one moment`.
110
+2. Require a full-bleed environment from corner to corner.
111
+3. Use one simple, visible physical action.
112
+4. Name exactly which characters are visible.
113
+5. Put stable hair, face, and wardrobe blocks before environmental detail.
114
+6. Ban borders, margins, vignettes, dioramas, posters, split screens, panels,
115
+ storyboards, duplicated scenes, text, and watermarks in the model adapter.
116
+7. Do not state a height relationship.
117
+8. Treat pendant, scarf, and satchel continuity as secondary until references
118
+ are conditioned directly.
119
+
120
+## Style test
121
+
122
+`Mia-ReadingNook-v007_00001_.png` became the working style target. Ten new
123
+images reused one verbatim style block across different characters,
124
+environments, lighting, and actions.
125
+
126
+Closest text-only matches:
127
+
128
+- `Style-Mia-Breakfast-v009_00001_.png`
129
+- `Style-Mia-FlowerShop-v009_00001_.png`
130
+- `Style-Henry-Attic-v009_00001_.png`
131
+- `Style-Pair-Cookies-v009_00001_.png`
132
+- `Style-Pair-CabinMap-v009_00001_.png`
133
+- `Style-Pair-Porch-v010_00001_.png`
134
+
135
+Observed drift:
136
+
137
+- One frame moved toward glossy 3D rendering.
138
+- Cool or rainy environments changed the palette and line treatment.
139
+- Some frames became flatter or more graphic despite identical style text.
140
+
141
+**Conclusion:** fixed text can create a recognizable style family, but it does
142
+not lock the exact Reading Nook rendering. The next style experiment should
143
+condition on an approved image reference or test a dedicated style LoRA.
144
+
145
+## Wan 2.2 14B Turbo I2V tests
146
+
147
+Two full-frame stills were uploaded to separate local ComfyUI instances and
148
+animated concurrently.
149
+
150
+| Test | Source | Motion intent | Prompt ID | Result |
151
+| --- | --- | --- | --- | --- |
152
+| Snow Village | `Pair-SnowLanterns-v005-FullFrame_00001_.png` | Slow forward walk, scarf movement, lantern sway and flicker, falling snow, gentle backward track | `f5036d1f-362f-4bf8-86fc-dc94b5a8098c` | Technical success |
153
+| Pip Bakery | `Pip-Bakery-v005-FullFrame_00001_.png` | One restrained kneading cycle, blink, cloth settling, oven flicker, slight push-in | `c178465d-fc56-4c0f-b6a4-ac7c916be7d4` | Technical success |
154
+
155
+### I2V configuration
156
+
157
+| Field | Value |
158
+| --- | --- |
159
+| Models | Wan 2.2 I2V 14B FP8 high-noise and low-noise |
160
+| Acceleration | `wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors` and low-noise counterpart |
161
+| Steps / split | 4 / 2 |
162
+| CFG | 1 |
163
+| Sampler / scheduler | Euler / simple |
164
+| Output | 480 × 640, 5 seconds, 16 fps |
165
+| Runtime | Approximately 2 minutes 20 seconds per test while running concurrently |
166
+
167
+The first tests prove that the workflow executes successfully on both A4000s.
168
+They do not yet approve motion quality, identity preservation, or Turbo as the
169
+final production setting.
170
+
171
+## Decisions and next experiments
172
+
173
+### Approved
174
+
175
+- Character names: Mia and Henry.
176
+- Do not prompt a height relationship.
177
+- Reject solid edges, blank margins, vignettes, dioramas, split panels, and
178
+ duplicated scenes.
179
+
180
+### Candidate
181
+
182
+- The revised everyday Mia and Henry appearances.
183
+- Mia Reading Nook as the style target.
184
+- The current LightX2V four-step Wan I2V workflow.
185
+
186
+### Next
187
+
188
+1. Select exact Mia, Henry, and pair references.
189
+2. Build the required turnaround, expression, action, seated, and neutral
190
+ two-shot pack.
191
+3. Test reference-conditioned character generation.
192
+4. Test image-reference style matching or a dedicated style LoRA.
193
+5. Compare the same approved first frame in Wan Turbo and the full 20-step
194
+ workflow.
195
+6. Review face, hands, accessory continuity, texture crawl, flicker, and motion
196
+ quality before selecting an I2V workflow.
Decision-Log.md
... ...
@@ -4,6 +4,12 @@ Use this page for decisions that change production canon. Newest entries go firs
4 4
5 5
| Date | Area | Decision | Status | Affected pages |
6 6
| --- | --- | --- | --- | --- |
7
+| 2026-07-27 | Character names | Replace the working names Lumi and Pip with **Mia and Henry** | Approved | [Cast](/Cast), [Character, Style, and I2V Tests](/Character-Style-and-I2V-Tests-2026-07-27) |
8
+| 2026-07-27 | Character direction | Test Mia and Henry as familiar, everyday children with simpler hair, clothing, and facial anchors; keep accessories secondary until reference-guided continuity is proven | Candidate visual direction | [Cast](/Cast), [Character, Style, and I2V Tests](/Character-Style-and-I2V-Tests-2026-07-27) |
9
+| 2026-07-27 | Character scale prompting | Do not state a height relationship between Mia and Henry in image prompts | Approved prompting rule | [Cast](/Cast), [Prompt Architecture](/Prompt-Architecture) |
10
+| 2026-07-27 | Image composition | Require a single continuous full-bleed frame; reject solid borders, blank margins, vignettes, isolated dioramas, split panels, and duplicated scenes | Approved prompting and review rule | [Prompt Architecture](/Prompt-Architecture), [Character, Style, and I2V Tests](/Character-Style-and-I2V-Tests-2026-07-27) |
11
+| 2026-07-27 | Style direction | Use the Mia Reading Nook image as the current style target; treat text-only style matching as a family resemblance rather than an exact lock | Candidate | [Prompt Architecture](/Prompt-Architecture), [Character, Style, and I2V Tests](/Character-Style-and-I2V-Tests-2026-07-27) |
12
+| 2026-07-27 | Wan I2V acceleration | Continue evaluating Wan 2.2 14B I2V with the LightX2V high- and low-noise four-step LoRAs; the first two local tests proved technical execution only | Candidate workflow | [Workflow Registry](/Workflow-Registry), [Character, Style, and I2V Tests](/Character-Style-and-I2V-Tests-2026-07-27) |
7 13
| 2026-07-26 | Season 01 music | Use an acoustic-storybook-adventure album identity, six rotating song families, fixed series signatures, and varied episode composition across the A–Z music map | Approved direction | [Season 01 — Music Bible](/Season-01---Music-Bible), [Music and Voices](/Music-and-Voices) |
8 14
| 2026-07-26 | Shot timing | Import the finished Suno song into Palmier, transcribe it with Whisper, review the exact timecodes, and derive shot ranges before Anima storyboarding | Approved | [Music and Voices](/Music-and-Voices), [Visual Pipeline](/Visual-Pipeline), [Episode Template](/Episode-Template) |
9 15
| 2026-07-26 | Pipeline order | Suno is always the first media-generation stage; lock its selected audio and timing map before Anima storyboarding, then use Wan and finish in Palmier | Approved | [Music and Voices](/Music-and-Voices), [Visual Pipeline](/Visual-Pipeline), [Little Lanterns Format](/Little-Lanterns-Format) |
Prompt-Architecture.md
... ...
@@ -26,12 +26,13 @@ style:
26 26
palette: warm lantern gold against teal-blue shadows
27 27
28 28
cast:
29
- visible: [Lumi]
30
- lumi:
31
- hair: deep navy twin buns
32
- eyes: large teal
33
- wardrobe: teal and cream with warm-gold details
34
- signature: small lantern pendant
29
+ visible: [Mia]
30
+ mia:
31
+ hair: chestnut-brown chin-length bob with one small teal clip
32
+ eyes: warm brown
33
+ face: light freckles
34
+ wardrobe: teal zip jacket, cream T-shirt, denim pants
35
+ signature: tiny round lantern pendant
35 36
36 37
location:
37 38
name: Lantern Club room
... ...
@@ -43,10 +44,10 @@ shot:
43 44
lens_intent: intimate portrait compression
44 45
camera_height: eye level
45 46
subject_scale: face fills most of frame
46
- composition: Lumi alone; clean eyeline space camera-right
47
+ composition: Mia alone; clean eyeline space camera-right
47 48
action: she notices the letter glowing
48 49
emotion: quiet wonder
49
- continuity: pendant visible; background softly recognizable
50
+ continuity: teal clip visible; background softly recognizable
50 51
51 52
output:
52 53
aspect_ratio: 3:4
... ...
@@ -57,7 +58,7 @@ output:
57 58
58 59
1. Include only visible character features in the final shot prompt.
59 60
2. Put shot framing and subject scale near the beginning and reinforce them in composition.
60
-3. State exactly who is visible; “Lumi alone” is stronger than omitting Pip.
61
+3. State exactly who is visible; “Mia alone” is stronger than omitting Henry.
61 62
4. Use stable location anchors, but vary camera position and focal depth.
62 63
5. Preserve style language verbatim across a sequence.
63 64
6. Keep negative constraints in the model adapter, not scattered through story fields.
... ...
@@ -68,3 +69,35 @@ output:
68 69
Perfect pixel identity is unrealistic across independent generations. Strong,
69 70
orthogonal visual anchors—silhouette, hair shape, palette, signature object, and
70 71
movement—make small variance tolerable while preserving recognition.
72
+
73
+## Composition rules learned from Anima tests
74
+
75
+1. Require a **single continuous camera image, one frame, one moment**.
76
+2. Require the environment to fill every pixel from corner to corner.
77
+3. Reject solid-color borders, blank margins, vignettes, circular frames,
78
+ isolated dioramas, poster layouts, split screens, comic panels, storyboards,
79
+ and duplicated scenes.
80
+4. Prefer one simple physical action. Multiple sequential actions increase the
81
+ chance that Anima will create two panels or a before-and-after layout.
82
+5. State subject scale and camera geometry, but do not state a height
83
+ relationship between Mia and Henry.
84
+6. Treat small accessories as secondary continuity details until
85
+ reference-guided generation proves them stable.
86
+
87
+## Current style target
88
+
89
+The current target is the rendering family represented by
90
+`Mia-ReadingNook-v007_00001_.png`:
91
+
92
+> Warm cinematic storybook animation matching a polished contemporary
93
+> children's picture book: soft amber backlight and gentle window glow, rich
94
+> but muted teal, cream, rust, navy, and honey-brown palette, rounded natural
95
+> child faces, medium-large expressive brown eyes, fine dark-brown linework,
96
+> softly shaded 2.5D forms, subtle fabric, hair, paper, and wood texture,
97
+> detailed lived-in environment, gentle atmospheric depth, balanced full-bleed
98
+> composition.
99
+
100
+Keeping that text unchanged produced a coherent style family but not an exact
101
+lock. Exact matching will require a validated style-reference workflow,
102
+reference conditioning, or a tested style LoRA. See
103
+[Character, Style, and I2V Tests — 2026-07-27](/Character-Style-and-I2V-Tests-2026-07-27).
Workflow-Registry.md
... ...
@@ -8,9 +8,11 @@ review link whenever a workflow changes.
8 8
| Music and narration | Create and select the finished episode song and narration | Suno workflow | Exact prompt version TBD | Approved first stage |
9 9
| Audio timeline ingest | Place the selected Suno song on the authoritative episode timeline | Palmier | Exact project/template version TBD | Approved second stage |
10 10
| Song transcription and timing | Produce and review the timecoded transcript used to define shots | Whisper | Exact model/workflow version TBD | Approved third stage |
11
-| Character reference pack | Turnaround, expressions, pair scale | Local ComfyUI | TBD | Draft |
11
+| Character reference pack | Turnaround, expressions, and shared neutral two-shot for Mia and Henry | Local ComfyUI / Anima | `EXP-TLC-001`; exact reference-conditioned workflow TBD | Draft |
12
+| Character and environment still test | Test visible character anchors, full-bleed composition, and ordinary environments at 3:4 | Local ComfyUI / Anima | `anima-base-v1.0`; 864 × 1152; 30 steps; CFG 4; `er_sde` / `simple` | Verified experiment |
13
+| Reading Nook style-family test | Reuse one verbatim text style block across ten different scenes | Local ComfyUI / Anima | `EXP-TLC-001`; exact reference conditioning TBD | Text-only family verified; exact lock unproven |
12 14
| 12-panel first-frame storyboard | Interpret the reviewed Whisper/Palmier timing with consistent cast, location, varied coverage, and one approved first frame per shot | Local ComfyUI / Anima | Mary–Joseph V1 is the current learning reference | Approved fourth stage |
13
-| Image-to-video shot | Animate an approved Anima first frame to its assigned Palmier time range without redesigning it | Local ComfyUI / Wan 2.2 14B | Exact workflow path and version TBD | Approved fifth stage |
15
+| Image-to-video shot | Animate an approved Anima first frame to its assigned Palmier time range without redesigning it | Local ComfyUI / Wan 2.2 14B | FP8 high/low-noise models; LightX2V four-step LoRAs; first local tests at 480 × 640, 5 s, 16 fps | Approved fifth stage; Turbo quality comparison pending |
14 16
| Episode finishing | Assemble, time, revise, and finalize every selected media element | Palmier | Exact project/template version TBD | Approved interim direction |
15 17
| Asset review | Contact sheets and approved output collections | SmartGallery | TBD | Candidate |
16 18