nodes/olivv-cs-ComfyUI-FunPack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

MiniMax H3-Image saves a PNG picture; Length and FPS are not offered and are set to 1

Browse Source 
A pipeline with no length or rate input hides those controls in the Project and Scene panels. A preset can carry project numbers it goes with.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
digital-garbage
2026-11-11 03:27:51 +02:00
co-authored by Claude Sonnet 5.5
parent 98a5755920
commit 30d1aa14a9
11 changed files with 66 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+1 -1
View File
@@ -4,7 +4,7 @@ What is new in each version, in plain words. Newest first.
## [Unreleased]
- **MiniMax H3-Image** (experimental): a new "Start from" pipeline, MiniMax H3-Image · Text to Image and Edit. Write a prompt for a picture, or add reference pictures and refer to them as `<Picture 1>`, `<Picture 2>`, ... in the prompt (the first is the picture to edit; up to 16). The picture is held for the clip's length like other image models. You need the model, its VAE and its text encoder files; the model's licence excludes the EU, the UK, South Korea and the United States.
- **MiniMax H3-Image** (experimental): a new "Start from" pipeline, MiniMax H3-Image · Text to Image and Edit. Write a prompt for a picture, or add reference pictures and refer to them as `<Picture 1>`, `<Picture 2>`, ... in the prompt (the first is the picture to edit; up to 16). The result is a PNG picture, not a clip, and Length and FPS are not offered for it (both are set to 1). You need the model, its VAE and its text encoder files; the model's licence excludes the EU, the UK, South Korea and the United States.
- **Noise merging is now called Step merging**, because most of its modes merge the model's answers at a step, not noise. Settings are kept.
- **Alternating prompts** (Settings > general): write `[dance|zoom]` in a prompt, or `DANCE|ZOOM` between two shortcuts, for two options, A and B. "Main prompt reads" sets what the prompt says by default: "A while B" (default), "A B", only A or only B. Switch on "Read A and B separately" (off by default) to also assign steps: "Steps that read only A" and "Steps that read only B" give each step you name its own option, with no extra model pass. A `|` next to anything that is not a shortcut is left as typed. Two options per bracket.
- **Simultaneous actions** (Step merging, needs "Read A and B separately"): with `[A|B]` in the prompt, each merged step is answered once on A and once on B from the same picture, and the two answers are mixed at your strength. "Second pass new noise" sets how much of the noise B starts from is new (0 keeps the same noise for both). Strength 0 reads only A, and strength 1 with new noise 1 reads only B, each with a single pass. Steps named in Step merging win over the per-step lists. Costs one extra model pass per merged step.
app/ui/features/inspector/scene.js
+10 -9
View File
@@ -2,7 +2,7 @@
import { composer as c } from "../../composer/composer.js";
import { namesOf, pick } from "./mediapick.js";
import { effFrames, effFps, genUnitId, isSubclip, seconds, snapFrames } from "../../shell/scenes.js";
import { gridOf } from "../../shell/frame_grid.js";
import { gridOf, offersLength } from "../../shell/frame_grid.js";
import { restoreToPlan } from "../../shell/edits.js";
const MODES = [{ value: "project", label: "Project default" }, { value: "timeline", label: "Timeline trim" }, { value: "custom", label: "Custom" }];
@@ -37,7 +37,7 @@ function drift(sc, open) {
export const card = (...children) => c.card.default({ children: children.flat().filter(Boolean) });
export function sceneRows(p, app) {
const sc = p.selected, open = p.project;
const sc = p.selected, open = p.project, slots = app.pipeline && (app.pipeline.view || app.pipeline.slots)();
if (!sc) return [c.emptyState.default({ icon: "▭", title: "No scene", hint: "Add one on the timeline." })];
const cuts = p.scenes.filter((s) => genUnitId(s) === genUnitId(sc)).length > 1;
return [
@@ -49,13 +49,14 @@ export function sceneRows(p, app) {
card(mediaField(app, p, sc, { label: "Aspect ratio source", hint: "The scene's picture. Its shape sets the aspect ratio; the project's Width and Height set the size.",
key: "source_image", multiple: false, kinds: ["image", "video"], empty: "— choose aspect ratio source —" }),
mediaField(app, p, sc, { label: "References", key: "references", multiple: true, kinds: ["image", "video", "audio"], empty: "+ Add reference" })),
c.label.section({ text: "Length" }),
card(c.field.row({ fields: [
lengthField(p, sc, "Frames", "frames_mode", "frames", effFrames(sc, open), gridOf(app.pipeline && (app.pipeline.view || app.pipeline.slots)())),
lengthField(p, sc, "FPS", "fps_mode", "fps", effFps(sc, open)),
] }),
drift(sc, open),
c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") })),
...(offersLength(slots) ? [c.label.section({ text: "Length" }),
card(c.field.row({ fields: [
lengthField(p, sc, "Frames", "frames_mode", "frames", effFrames(sc, open), gridOf(slots)),
lengthField(p, sc, "FPS", "fps_mode", "fps", effFps(sc, open)),
] }),
drift(sc, open),
c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") }))] : [
card(c.button.md({ label: "Generate this scene", tone: "primary", onClick: () => app.say("generate.scene") }))]),
card(c.checkboxRow.default({ label: "Exclude from full generation", hint: "Generate leaves this scene out.", checked: Boolean(sc.excluded), onChange: (v) => p.setScene(sc.id, "excluded", v) })),
c.banner.info({ text: "Made with the project's pipeline and Engine settings", action: { label: "Engine settings →", onClick: () => app.openSettings && app.openSettings("engine") } }),
];
app/ui/features/settings/models.js
+4 -1
View File
@@ -228,7 +228,10 @@ export const models = (app) => function mount() {
const pick = found.find((p) => p.id === id);
if (!pick || !(await confirm("Replace pipeline", `Replace the whole pipeline with “${pick.title}”? Every file and value you set here is cleared. Revert puts it back until you leave this page.`, "danger"))) return;
const r = await ps.restore({ slots: pick.slots, removed: [], unwired: {} });
if (!r.refused.length) Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v)); // project settings the preset goes with (render steps)
if (!r.refused.length) { // project settings the preset goes with (render steps, a picture model's one frame)
Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v));
Object.entries(pick.project || {}).forEach(([k, v]) => app.project.setField(k, v));
}
specs = {}; await describe(); say(r.refused.length ? `Not changed: ${r.refused[0]}` : "");
}
async function importWorkflow() {
app/ui/features/wizard/index.js
+1 -1
View File
@@ -28,7 +28,7 @@ export default {
const pick = presets.find((p) => p.id === ctx.preset);
try {
await app.project.newProject((ctx.name || "").trim() || "Untitled montage");
if (pick) { const r = await app.pipeline.restore({ slots: pick.slots, removed: [], unwired: {} }); if (r.refused.length) c.toast.warn({ text: `The model's pipeline was not loaded: ${r.refused[0]}` }); else Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v)); }
if (pick) { const r = await app.pipeline.restore({ slots: pick.slots, removed: [], unwired: {} }); if (r.refused.length) c.toast.warn({ text: `The model's pipeline was not loaded: ${r.refused[0]}` }); else { Object.entries(pick.prefs || {}).forEach(([k, v]) => app.project.setPref(k, v)); Object.entries(pick.project || {}).forEach(([k, v]) => app.project.setField(k, v)); } }
} catch (err) { busy = false; return c.toast.warn({ text: `Could not make the project: ${err.message}` }); }
busy = false;
close();
app/ui/shell/frame_grid.js
+4
View File
@@ -21,3 +21,7 @@ export async function learnGrid(api, slots) {
try { Object.assign(specs, (await api.describeNodes(missing)).nodes || {}); } catch { missing.forEach((n) => { specs[n] = null; }); }
return true;
}
/** Whether the pipeline takes a length or a rate from the project at all. A model that makes a picture holds it for a fixed time, so
* there is nothing to set: the Length and FPS controls are not offered. A pipeline not known yet offers them. */
export const offersLength = (slots) => !(slots && slots.length) || slots.some((s) => (s.roles || []).some((r) => r.drives === "frames" || r.drives === "fps"));
app/ui/shell/frame_grid.test.js
+12 -1
View File
@@ -1,6 +1,6 @@
import test from "node:test";
import assert from "node:assert/strict";
import { gridOf, learnGrid } from "./frame_grid.js";
import { gridOf, learnGrid, offersLength } from "./frame_grid.js";
test("the frame grid is read from the model's own length input: H3 17k+5, LTX 8k+1, a plain count none", async () => {
const slots = [{ id: "r2v", node: "H3Node", roles: [{ at: "project.video", input: "length", drives: "frames" }] },
@@ -14,3 +14,14 @@ test("the frame grid is read from the model's own length input: H3 17k+5, LTX 8k
await learnGrid({ describeNodes: async () => ({ nodes: { LtxNode: { widgets: [{ name: "length", min: 1, step: 8 }] } } }) }, ltx);
assert.deepEqual(gridOf(ltx), { step: 8, base: 1, fps: null });
});
test("Length and FPS are offered when a pipeline takes them, or is not known yet, and not for one that holds a picture for a fixed time", () => {
const length = { id: "r", roles: [{ at: "project.video", input: "length", drives: "frames" }] };
const rate = { id: "s", roles: [{ at: "project.video", input: "fps", drives: "fps" }] };
const size = { id: "a", roles: [{ at: "project.video", input: "width" }] };
assert.equal(offersLength([size, length]), true);
assert.equal(offersLength([size, rate]), true);
assert.equal(offersLength([size, { id: "x" }]), false);
assert.equal(offersLength([]), true);
assert.equal(offersLength(undefined), true);
});
core/routes.py
+3 -1
View File
@@ -328,7 +328,9 @@ def register(routes, prefix=None):
if isinstance(preset, dict) and preset.get("id") and preset.get("slots"):
found.append({"id": preset["id"], "title": preset.get("title") or preset["id"],
"module": spec.id, "slots": preset["slots"],
"prefs": preset.get("prefs") if isinstance(preset.get("prefs"), dict) else {}})
"prefs": preset.get("prefs") if isinstance(preset.get("prefs"), dict) else {},
# project numbers the pipeline goes with (a picture model: 1 frame at 1 fps)
"project": {k: v for k, v in (preset.get("project") or {}).items() if k in ("num_frames_per_scene", "frame_rate") and isinstance(v, int)}})
return web.json_response({"presets": found})
@routes.post(P + "/api/pipeline/import")
core/tests/test_routes_pipeline.py
+7
View File
@@ -532,6 +532,13 @@ def test_a_module_provided_preset_is_offered(server, registered):
assert preset["slots"]
def test_a_picture_model_preset_carries_the_project_numbers_it_goes_with(server, registered):
_status, body = _request(server, "GET", "/fun/api/pipeline/presets")
image = next(p for p in body["presets"] if p["id"] == "minimax_h3_image")
assert image["project"] == {"num_frames_per_scene": 1, "frame_rate": 1}
assert next(p for p in body["presets"] if p["id"] == "minimax_h3_reference_to_video")["project"] == {} # a video model changes none
def test_a_comfyui_workflow_imports_into_slots_that_build_a_real_graph(server, registered):
"""Over HTTP, against ComfyUI's own nodes: the API export of a plain text-to-image graph."""
from core import graph
docs/how-to-start.md
+1 -1
View File
@@ -24,7 +24,7 @@ Start ComfyUI as usual and open `/fun/` on the same address, for example `http:/
|---|---|
| MiniMax H3 · Reference to Video, MiniMax H3 · Fast & Simple R2V | The model FunUI is built and tested around. |
| Stable Diffusion 1.5 · Text to Image (held as a clip) | Experimental. |
| MiniMax H3-Image · Text to Image and Edit (held as a clip) | Experimental. Add pictures as references: the first is the picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...), up to 16. Uses 6 steps at CFG 1; Width and Height set the size (about 4 megapixels is a good start). |
| MiniMax H3-Image · Text to Image and Edit | Experimental. Add pictures as references: the first is the picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...), up to 16. The result is a picture file (PNG), and Length and FPS are not offered (both are set to 1). Uses 6 steps at CFG 1; Width and Height set the size (about 4 megapixels is a good start). |
| LTX 2.3 · Image to Video, from References, LTX 2.3 · Video to Video, direct | Tested on a real model, with LoRAs, Step merging and Attention explorer. Other enhancements are not tested on LTX yet; please report problems in [Issues](https://github.com/digital-garbage/FunUI/issues). |
## 4. Pick your files
modules/models/minimax_h3_image/__init__.py
+9 -11
View File
@@ -5,9 +5,8 @@ the 16x latent grid, has no sound stream and reads at most 16 pictures: the firs
picture to edit (`<Picture 1>` in the prompt), the others are references (`<Picture 2>`, ...).
It samples in 6 steps at CFG 1 and works best around 4 to 16 megapixels.
Like every image model here, its picture is held for the clip's length, so the editor gets a clip
made the same way as any other. A held 16-megapixel frame is about 48 MB in memory: keep the
length short at large sizes.
The result is a picture file (PNG), not a clip. The pipeline offers no Length or FPS; the preset sets
the project's Length and FPS to 1, so a scene made with it counts as one second on the timeline.
The checkpoint is recognised by its header: the packed-DiT keys of H3 with no audio projection.
Everything the model needs at run time (the image latent, the picture labels, the sigma grid
@@ -31,6 +30,9 @@ _ABSENT = "audio_patch_proj.weight"
#: How many pictures the model reads (the one to edit and its references).
MAX_REFERENCES = 16
#: The project numbers the preset sets: one frame at one frame per second.
PROJECT = {"num_frames_per_scene": 1, "frame_rate": 1}
def detect(keys) -> bool:
keyset = set(keys)
@@ -52,7 +54,6 @@ def probe_traits(keys) -> list:
def text_to_image():
video = {"at": "project.video", "input": "fps", "label": "FPS", "drives": "fps"}
slots = [
{"id": "model", "group": "Loaders", "node": "FunUIDiffusionModelLoader", "inputs": {
"weight_dtype": "default", "compute_dtype": "default", "attention": "default"}},
@@ -94,11 +95,8 @@ def text_to_image():
{"id": "decode", "group": "Render", "node": "FunUIDecode",
"inputs": {"samples": ["sampler", 0], "vae": ["vae", 0], "model": ["model", 0]}},
{"id": "hold", "group": "Render", "node": "RepeatImageBatch",
"roles": [{"at": "project.video", "input": "amount", "label": "Length", "drives": "frames"}],
"inputs": {"image": ["decode", 0], "amount": 1}},
{"id": "save", "group": "Render", "node": "FunUISaveVideo", "roles": [video],
"inputs": {"images": ["hold", 0], "fps": 24.0, "filename_prefix": "FunUI", "settings": ["settings", 0]}},
{"id": "save", "group": "Render", "node": "SaveImage",
"inputs": {"images": ["decode", 0], "filename_prefix": "FunUI"}},
]
for n in range(1, MAX_REFERENCES + 1):
# Not wired here: an unused picture input is simply absent. The app wires the ones a scene has, per run.
@@ -111,8 +109,8 @@ def text_to_image():
def presets():
return [{"id": "minimax_h3_image", "title": "MiniMax H3-Image · Text to Image and Edit (held as a clip)",
"slots": text_to_image()}]
return [{"id": "minimax_h3_image", "title": "MiniMax H3-Image · Text to Image and Edit",
"slots": text_to_image(), "project": PROJECT}]
PROVIDES = {"detect": detect, "probe_traits": probe_traits, "pipeline_presets": presets}
modules/models/minimax_h3_image/tests/test_h3_image.py
+14
View File
@@ -54,3 +54,17 @@ def test_the_preset_builds_a_real_graph_and_is_offered(registered):
assert refs[0]["roles"][0]["wireTo"] == {"slot": "cond", "input": "images.image_1"} # image_1 is the picture to edit
sampler = next(s for s in slots if s["id"] == "sampler")["inputs"]
assert (sampler["steps"], sampler["cfg"], sampler["sampler_name"], sampler["scheduler"]) == (6, 1.0, "euler", "simple")
def test_it_saves_a_picture_and_offers_no_length_or_fps(registered):
from core import graph, routes
from modules.models.minimax_h3_image import PROJECT, presets, text_to_image
slots = {s["id"]: s for s in text_to_image()}
assert not [r for s in slots.values() for r in s.get("roles") or [] if r.get("drives") in ("frames", "fps")]
assert slots["save"]["node"] == "SaveImage" and "hold" not in slots # a PNG, not a clip
assert PROJECT == {"num_frames_per_scene": 1, "frame_rate": 1} and presets()[0]["project"] == PROJECT
prompt, _problems = graph.build(list(slots.values()))
assert [n["class_type"] for n in prompt.values()].count("SaveImage") == 1
served = [p for _s, make in routes.modules().providers("pipeline_presets") for p in make() if p["id"] == "minimax_h3_image"]
assert served and served[0]["project"] == PROJECT