akhaliq HF Staff commited on
Commit
8e88a7e
·
verified ·
1 Parent(s): b2540ad

chained storyboard GIF demos (lovis93 style)

Browse files
Files changed (1) hide show
  1. README.md +27 -10
README.md CHANGED
@@ -66,24 +66,41 @@ Next Scene: The camera pans right, revealing the fleet massing behind the ridge.
66
  - Chain generations (output → input) to build sequential storyboards.
67
  - If the output drifts toward copying the input, lower the LoRA strength toward 0.7.
68
 
69
- ## Real photos (out-of-distribution)
70
 
71
- The LoRA was trained only on film frames (Sintel + Tears of Steel) — here it is on
72
- everyday CC photos it never saw, step-1,500 checkpoint, strength 0.8, 28 steps, seed 42,
73
- same prompt for both:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
74
 
75
  > *Next Scene: The camera pulls back from the close-up to a wide shot, revealing the sunlit meadow around the subject, soft morning haze.*
76
 
77
- **Golden retriever — input:**
78
  ![dog input](examples/input_dog.jpg)
79
-
80
- **Golden retriever — next scene:** identity holds, framing pulls back to the meadow
81
  ![dog next scene](examples/real_dog_0.jpg)
82
 
83
- **Cat — input:**
84
  ![cat input](examples/input_cat.jpg)
85
-
86
- **Cat — next scene:**
87
  ![cat next scene](examples/real_cat_0.jpg)
88
 
89
  Photo credits (CC): [Golden Retriever Carlos](https://commons.wikimedia.org/wiki/File:Golden_Retriever_Carlos_(10581910556).jpg),
 
66
  - Chain generations (output → input) to build sequential storyboards.
67
  - If the output drifts toward copying the input, lower the LoRA strength toward 0.7.
68
 
69
+ ## 🎬 Chained storyboards (the point of this LoRA)
70
 
71
+ Each GIF is **one starting photo + 3 chained generations** — every output feeds back as
72
+ the next input, with a different director's instruction per hop. No video model involved.
73
+ Step-1,500 checkpoint, strength 0.8, 28 steps per hop.
74
+
75
+ **Chain 1 — golden retriever (real CC photo in):**
76
+ ![dog storyboard](examples/chain_dog.gif)
77
+
78
+ 1. *"Next Scene: The camera pulls back to a wide shot, revealing the sunlit meadow…"*
79
+ 2. *"Next Scene: The camera pans right as the subject bounds toward the treeline, long shadows…"*
80
+ 3. *"Next Scene: The camera pushes in on the subject as golden-hour light breaks through the trees, lens flare blooming."*
81
+
82
+ **Chain 2 — cat (real CC photo in):**
83
+ ![cat storyboard](examples/chain_cat.gif)
84
+
85
+ 1. *"Next Scene: The camera pulls back, revealing a rustic garden patio…"*
86
+ 2. *"Next Scene: Cut to a low-angle shot as the subject steps onto a stone ledge, backlit…"*
87
+ 3. *"Next Scene: The camera tracks forward through drifting leaves, closing in as dusk settles."*
88
+
89
+ Identity, palette and world hold across 3 hops; the camera follows the instruction each time.
90
+ Individual frames: `examples/chain_{dog,cat}_{1,2,3}.jpg`.
91
+
92
+ ## Real photos (single-hop)
93
+
94
+ Same real-photo inputs, one generation, step-1,500 checkpoint, strength 0.8:
95
 
96
  > *Next Scene: The camera pulls back from the close-up to a wide shot, revealing the sunlit meadow around the subject, soft morning haze.*
97
 
98
+ **Golden retriever — input → next scene:**
99
  ![dog input](examples/input_dog.jpg)
 
 
100
  ![dog next scene](examples/real_dog_0.jpg)
101
 
102
+ **Cat — input → next scene:**
103
  ![cat input](examples/input_cat.jpg)
 
 
104
  ![cat next scene](examples/real_cat_0.jpg)
105
 
106
  Photo credits (CC): [Golden Retriever Carlos](https://commons.wikimedia.org/wiki/File:Golden_Retriever_Carlos_(10581910556).jpg),