docs: head-to-head vs Wan2.2-Animate, with clips

.gitattributes

@@ -42,3 +42,13 @@ examples/media/compare-horns.mp4 filter=lfs diff=lfs merge=lfs -text
 examples/media/compare-costume.mp4 filter=lfs diff=lfs merge=lfs -text
 examples/media/compare-prop.mp4 filter=lfs diff=lfs merge=lfs -text
 examples/media/compare-style.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-airliner.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-anime-duo.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-claymation.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-corgi.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-panda.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-penguin.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-photoreal-duo.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-robot.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-wushu-animals.mp4 filter=lfs diff=lfs merge=lfs -text
+examples/media/swap-wushu-robots.mp4 filter=lfs diff=lfs merge=lfs -text

README.md

@@ -101,6 +101,77 @@ These three are selected. Across the 24 clips we ran head to head, Wan came out
 tight talking-head shot, where its explicit face channel tracks expression that we render flat. That
 is the same weakness listed under Limitations below. Roughly six were a wash.
 
+## What you can paint in
+
+The model is never told what the new character is. It sees one edited frame and has to carry it
+through five seconds of motion it did not choreograph — so the question is how far from a person
+that character can get. Three panels each: **painted reference · driving video · this model.** All
+ten are single attempts at both stages: one paint, one render, `--steps 4 --seed 42`, no best-of-N.
+
+### Animals
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-corgi.mp4"></video>
+
+A corgi in the driver's own tank top and trousers. The ears track head rotation independently —
+nothing in the driving clip has ears to copy.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-panda.mp4"></video>
+
+A giant panda in a denim jacket and ripped jeans. Clothing deforms onto fur without arguing about
+it, and the eye patches stay symmetric through the hands-over-face beat.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-penguin.mp4"></video>
+
+An emperor penguin in a red floral suit. The paint put the flippers where the arms were, and the
+render animates them *as* arms — they sweep on the arms' trajectories while keeping a flipper's shape.
+
+### Not humanoid
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-airliner.mp4"></video>
+
+The hardest case here. A man standing with his arms out becomes an airliner whose wings are his arms
+and whose landing gear are his legs, and that binding survives the whole take. The geometry is done
+in the paint — the wings placed exactly where the arms were — and the model only has to keep it.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-robot.mp4"></video>
+
+Specular metal has to relight itself as the body turns rather than just deform. The cable joints at
+the elbows and knees stay put instead of migrating.
+
+### Stylised
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-claymation.mp4"></video>
+
+A stop-motion plasticine character in front of a kitchen that stays photographic. The style boundary
+holds for the whole clip — fingerprints and seams on the figure, real appliances behind it.
+
+### Two characters, one edit
+
+The prompt has to bind each new character to a specific person — "the girl on the left", "the athlete
+on the right" — or the image editor swaps one and leaves the other. All four of these held.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-anime-duo.mp4"></video>
+
+Two guitarists become a cat and a corgi, drawn in the clip's own cel-shaded style, so the swap has to
+survive as line art rather than as texture. The fretting hands stay on the necks.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-photoreal-duo.mp4"></video>
+
+The same clip with photoreal fur in front of a hand-drawn night sky. Neither side pulls the other
+across: the background stays anime and the characters stay photographic.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-wushu-animals.mp4"></video>
+
+A panda and a tiger through a synchronised *duilian* routine, in a wide arena shot where each figure
+is a few dozen pixels tall — landscape framing and two subjects, both outside the training
+distribution. They stay distinct from each other and neither swaps back.
+
+<video controls muted loop playsinline width="100%" src="https://huggingface.co/Viggle/Viggle-Animate/resolve/main/examples/media/swap-wushu-robots.mp4"></video>
+
+Same clip, two robots that differ only in metal — gold on the left, chrome on the right. They stay
+told apart through the jumps and stance changes, which is what would fail if the pair were being
+treated as one region.
+
 ## What this is
 
 Two parts, both derived from [`MiniMaxAI/MiniMax-H3`](https://huggingface.co/MiniMaxAI/MiniMax-H3)'s
@@ -169,9 +240,22 @@ in **40–60 s**. That quantization path is not shipped here.
 - **More steps is not better.** `--steps 4` is the operating point, not a degraded `--steps 12`. It
   renders sharper than its own teacher — Laplacian variance 16–50 % higher — which reads as detail
   sometimes and over-sharpening other times, and worsens with more steps.
+- **The reference defines only the views it contains.** A character painted from behind has no
+  front, and when the driving clip rotates to reveal one the model fills it from its own prior
+  rather than from your character. On a figure skater who spins to face camera, a swapped-in cat is
+  a cat while her back is turned, a generic woman at the moment she faces us, and a cat again when
+  she turns away. Prefer driving clips that keep one side to camera, or paint a frame that shows the
+  view the clip will eventually reach.
+- **A silhouette far from the driver's survives as colour, not as geometry.** A LEGO minifigure on
+  that same skating clip kept its palette and its yellow claw hands, but the moulded plastic body
+  reverted to human anatomy. The airliner above works because the paint bound its wings to the arms
+  — geometry the *driving pose* can still explain.
 - **No facial-landmark channel**, so tight close-up dialogue is the weakest case — Wan drives the
-  face from explicit landmarks and is visibly more expressive there. Trained on one
-  subject filling a portrait frame: wide shots, crowds and landscape framing are out of distribution.
+  face from explicit landmarks and is visibly more expressive there.
+- **Out of distribution is not the same as broken.** Training was one subject filling a portrait
+  frame. The last four gallery clips are two subjects in landscape framing, including a wide arena
+  shot, and they hold — but that is untuned territory with no evaluation behind it, and crowds
+  remain untested.
 - The driving clip's audio is dropped at input; the model emits its own track, and the fixed prompt
   asks for silence.
 

examples/media/swap-airliner.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a7a31fd225e258633e208f13c18ff8523517834a9ea2c67b97ed060972becf3
+size 397277

examples/media/swap-anime-duo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:205700e55841a5fd22f517c6eb08c447068d0417a8e902bc91c863dede743d86
+size 108986

examples/media/swap-claymation.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd08b83d8a57e77408203710cf55eccdf8d1ee136c22123cc42ed8ac14434528
+size 286267

examples/media/swap-corgi.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5954aeab0360ac8c1c5c62b5d35935520ac0feffdb8d11b8c55a6ff84f9d9239
+size 272402

examples/media/swap-panda.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bec174f093bf553b39d1f022b6fedd46b9616a6ac30fa10a939753cf5e4a5150
+size 390485

examples/media/swap-penguin.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea2ccf8ce410331d194403903e2d1e836079b2d8fbafebf113265e52aafb3cd0
+size 346127

examples/media/swap-photoreal-duo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2d7c7fb36c5b2779176484b0a2fcb1735578ea3cad2fa0a2c0b057218881554
+size 108311

examples/media/swap-robot.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8147fb8f85eca14e372ecd26f6d7b4556c649f5c980e0dd514a582ac8619077
+size 194541

examples/media/swap-wushu-animals.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ed178d0c88053df75803d904bef26c83c81da78c33d575ec8f5897701b622a7
+size 204001

examples/media/swap-wushu-robots.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63cc5dfe297ca472b7c54a9ef32aaa575f84e6cc1088a42d01236e40888785b2
+size 201465