Add tested sampler presets for Krea and MOSS

README.md

@@ -104,6 +104,20 @@ KSampler, KSampler (Advanced) and custom sigmas in one node, built on `comfy.sam
 - **Step cache**: `step_cache` (off / auto / easycache / lazycache) wraps ComfyUI's own EasyCache and LazyCache, which skip model calls on steps where the output barely changes and work across architectures (including audio+video models). `auto` turns on easycache from 15 steps up; distilled 4-8 step runs should stay off. `cache_threshold` (higher = faster, softer), `cache_start` and `cache_end` set when it may skip. Measured on Anima, 30 steps, RTX 5080: easycache 0.1 skipped 15/30 steps (1.8x, similarity to uncached 0.991), 0.2 skipped 17 (2.05x, 0.978), 0.4 skipped 19 (2.4x, 0.971); lazycache 0.2 gave 2.05x at 0.988. Results vary by model.
 - **Presets**: save, apply and delete named presets (steps, cfg, sampler, scheduler, denoise, runs, step cache). They are stored in ComfyUI's user folder under `promptforge/sampler_presets.json`.
 
+Tested starting points at about 1.5 MP (identity tests used a source picture through the edit graphs; every row was rendered and checked):
+
+| Model | T2I | Edit |
+|---|---|---|
+| Krea 2 FinePorn V6.1 | 10 steps, cfg 1, `euler`/`simple` (or `er_sde`/`beta`) | 10 steps, cfg 1, `euler`/`simple`, denoise 1 |
+| Krea 2 Turbo Anime / ReaKrea2 Turbo | 8 steps, cfg 1, `res_multistep`/`simple` or `euler`/`simple` | 8 steps, cfg 1, same |
+| MOSS FinalCut int8 / FP8 | 25 steps, cfg 3, `er_sde`/`sgm_uniform` | 20 steps, cfg 2, `euler`/`normal`, denoise 1 |
+| MOSS Jib Mix | 20 steps, cfg 1, `euler`/`simple` | 20 steps, cfg 2, `euler`/`normal`, denoise 1 |
+| MOSS Dirty | 25 steps, cfg 2, `er_sde`/`simple` | 20 steps, cfg 2, `euler`/`normal`, denoise 1 |
+| MOSS PrimeLust fp16 / nvfp4 | 30 steps, cfg 5, `er_sde`/`normal` | 20 steps, cfg 2, `euler`/`normal`, denoise 1 |
+| with a DMD LoRA (lodestones 8-step, sft 4-step) | 8 steps, cfg 1, `er_sde`/`simple`, LoRA strength 1 | 8 steps, cfg 1, `er_sde`/`simple`, denoise 1 |
+
+DMD on PrimeLust needs 12 steps and cfg 1.5. MOSS edit keeps the face well at `image_strength` 1.5-2.0 and denoise 1 (it is reference conditioning, not img2img: a denoise below ~0.8 tiles the image). Krea edit holds identity strongly at 8-10 steps, cfg 1. Presets with these names are already in the sampler list.
+
 ### PromptForge Quantize Text Encoder (`PROMPTFORGE_QuantizeTextEncoder`)
 Turns a Qwen3.5 model folder (shards plus, if present, a separate `model-mtp*.safetensors`) into one quantized ComfyUI text encoder in `models/text_encoders`, in under a minute. Format `mixed` (default) is NVFP4 for the MLP and INT8 ConvRot for the attention layers, which ComfyUI runs reliably for long prompts; an all-NVFP4 file fails on long prompts. The vision tower, embeddings, output layer and the multi-token-prediction (MTP) head stay in bf16. Pick the file as the prompt writer: ComfyUI then writes with speculative decoding from the MTP head. A 9B model: 18 GB -> 9.6 GB, a prompt written in 8 to 10 s, against 18 s from the folder with transformers. Run the node once per model and refresh the lists (r).
 

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "ComfyUI-PromptForge"
 description = "ComfyUI prompt enhancement suite: model/config node with GPU-aware auto settings, prompt enhancer with a preset catalog, and a standalone config finder."
-version = "1.28.9"
+version = "1.28.10"
 license = { file = "LICENSE" }
 dependencies = [
     "transformers",

sampler_defaults.py

@@ -21,6 +21,7 @@ FAMILY = {
     "krea2": {"sampler_name": "res_multistep", "scheduler": "simple"},
     "minimax_h3": {"sampler_name": "euler", "scheduler": "simple"},
     "ltxv": {"scheduler": "ltxv"},
+    "moss_image": {"sampler_name": "er_sde", "scheduler": "simple", "steps": 20, "cfg": 3.0},
 }
 DISTILLED_HINT = re.compile(r"turbo|distill|lightning|hyper|lcm|schnell|dmd|\b[2-9]step|\b[1-9]\d?step", re.I)
 DISTILLED = {"steps": 8, "cfg": 1.0}