Skip to content

AI media harness for planning, generating, and editing locally

TypeScript

Update Slopfab to 0.2.0

Bittid01a4888811b
diff --git a/RELEASE_NOTES.md b/RELEASE_NOTES.md
index 908f689..77b35eb 100644
--- a/RELEASE_NOTES.md
+++ b/RELEASE_NOTES.md
@@ -10,6 +10,7 @@ Changes since 0.4.0.
## Fixes and improvements
+- Updated the bundled Slopfab runtime to 0.2.0, including a higher MiniMax H3 Image canvas limit of 20 MP.
- Fixed Extend rejecting editing drafts that already have a saved source image.
- YuE2 now defaults to six flow steps with DPM++ 2M. The previous Midpoint solver remains available in music's Advanced settings, and saved generator templates using the old 32-step default upgrade once to six steps.
- On first launch after updating, each recent project gets a one-time cleanup of orphaned generated media, latents and caches. Other projects receive the cleanup when first opened; files still referenced by the project are preserved.
diff --git a/docs/slopfab-compatibility.md b/docs/slopfab-compatibility.md
index 11e7efa..1d04ea5 100644
--- a/docs/slopfab-compatibility.md
+++ b/docs/slopfab-compatibility.md
@@ -1,9 +1,9 @@
# Slopfab compatibility
-## MiniMax H3 Image (API 1.52.0)
+## MiniMax H3 Image (SlopFab 0.2.0, API 1.52.0)
The bundled Windows and Linux runtimes use SlopFab commit
-`9ba04d8` with CUDA and Vulkan enabled and FFmpeg disabled.
+`c84f285` with CUDA and Vulkan enabled and FFmpeg disabled.
The **MiniMax H3 Image** template selects `h3_image`, Flash attention and the
fixed six-update Euler grid. Vulkan uses portable arithmetic. Its download
catalog uses [Comfy-Org/MiniMax-H3-Image](https://huggingface.co/Comfy-Org/MiniMax-H3-Image):
@@ -25,6 +25,12 @@ refmods, generation effects, CFG, negative prompts, H3 previews and saved latent
features are disabled for this family. Existing installations receive the
template once; removing it is respected.
+The updated runtime raises the image canvas ceiling to 20,971,520 pixels
+(20 Mi-pixels), accepting 5120×4096, 4096×5120 and 4480×4480. Dimensions must
+remain multiples of 32. Native integration tests check the new boundary on
+CUDA and Vulkan and reject 5120×4128. This update preserves API 1.52.0 and the
+existing worker protocol.
+
The new runtime counts H3 video steps as updates rather than grid points.
Slopus translates its existing H3 video and audio counts before submission,
preserving saved schedules, LoRA recipes and Single Shot anchoring. Image-only
diff --git a/lib/slopfab/libslopfab.so b/lib/slopfab/libslopfab.so
index 22e286d..2e788aa 100644
--- a/lib/slopfab/libslopfab.so
+++ b/lib/slopfab/libslopfab.so
@@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
-oid sha256:1ea799923865c3a62786b4ff4028260136a5949c632fcea57a052336cd4755bb
+oid sha256:0b5e89000ece8fc787279c2dd13a8b25021813c77fa5f91597123e4d6722530f
size 24314936
diff --git a/lib/slopfab/slopfab.dll b/lib/slopfab/slopfab.dll
index 0a94ee1..3816c50 100644
--- a/lib/slopfab/slopfab.dll
+++ b/lib/slopfab/slopfab.dll
@@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
-oid sha256:72b1c4849691a8fa8686fa2601a06f3e82849b62a38b3381551b53b020dcd74b
+oid sha256:2b27104891204a3af07e175adcbb4d62292cc86660d360d33d39e4d7a28d1af8
size 20917248
diff --git a/src-tauri/core/src/slopfab/tests.rs b/src-tauri/core/src/slopfab/tests.rs
index 196df2c..e491cda 100644
--- a/src-tauri/core/src/slopfab/tests.rs
+++ b/src-tauri/core/src/slopfab/tests.rs
@@ -2011,3 +2011,26 @@ fn h3_image_rejects_incompatible_controls_and_counts_the_edit_source() {
request.image_edit.as_mut().unwrap().edits[0].outpaint_langevin_steps = Some(5);
assert!(configuration.validate_inputs(&request).unwrap_err().contains("zero Langevin"));
}
+
+#[test]
+fn h3_image_supports_twenty_mipixel_canvases() {
+ let mut configuration = Configuration::from_settings(&BTreeMap::new());
+ configuration.model_type = "h3_image".into();
+ configuration.attention = "flash2";
+ let path = std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("../../lib/slopfab").join(super::DLL_FILE_NAME);
+ let api = ffi::Api::load(&path).unwrap();
+ for platform in [ComputePlatform::Cuda13, ComputePlatform::Vulkan] {
+ for (width, height, valid) in [(5120, 4096, true), (4096, 5120, true), (4480, 4480, true), (5120, 4128, false)] {
+ let request = GenerationRequest { still_image: true, frames: 1, steps: 6, prompt: "A landscape".into(),
+ canvas_width: width, canvas_height: height, ..Default::default() };
+ let handle = RequestHandle::new(&api).unwrap();
+ configure_request(&api, &handle, &request, &configuration, platform, RequestPurpose::Plan, &ReferenceVideos::default()).unwrap();
+ if valid {
+ let plan = api.resolve(&handle).unwrap();
+ assert_eq!((plan.canvas_width, plan.canvas_height, plan.aligned_frames, plan.num_model_evaluations), (width, height, 1, 6));
+ } else {
+ assert!(api.resolve(&handle).err().unwrap().contains("20 Mi-pixels"));
+ }
+ }
+ }
+}