Tested and tuned for Return Current. Drop these .json files into the app via Import workflow (PNG / JSON) — they also open cleanly in ComfyUI itself.
The four image workflows share the same surrealist botanical default prompt (each rendering its own model name as a glass emblem), and the Wan 2.2 video workflow's default prompt is built to animate their outputs — an end-to-end demo chain out of the box.
The most feature-rich workflow in the set. Leverages the IdeogramStudio node for structured prompt control.
- JSON-formatted prompt with style, composition, and scene description fields
- Aspect ratio grid (8 presets) with resolution override
- Quality presets: 🐇 Turbo (12 steps) · 🦅 Default (20) · 🐢 Quality (28)
- Color palette picker — define up to 12 hex colors that influence the output
- Width/height fine-tuning below aspect ratio
- Power Lora Loader with trigger keyword tracking
- Text → JSON and Image → JSON prompt generation via Tools pane (requires LM Studio nodes)
Fast, clean T2I workflow built around the new Krea 2 model.
- UNET model dropdown (subfolder organized)
- DPRandomGenerator prompt input with
{ x | y | z }random substitution syntax - ResolutionSelector with aspect ratio + megapixel control
- KSampler with full parameter access (seed, steps, cfg, sampler, scheduler, denoise)
- Power Lora Loader with trigger keyword tracking
SDXL-architecture workflow using UNETLoader.
- UNET model dropdown
- Prompt and negative prompt
- KSampler parameters
- Power Lora Loader with trigger keyword tracking
Standard SDXL checkpoint workflow. The baseline.
- Checkpoint model dropdown (subfolder organized)
- Prompt and negative prompt
- KSampler parameters (seed, steps, cfg, sampler, scheduler, denoise)
- Width/height controls
- Power Lora Loader with trigger keyword tracking
Two-stage HIGH/LOW I2V pipeline tuned for speed via Lightning LoRAs. The default prompt is designed to animate outputs from the other example workflows — generate a botanical, feed it straight in.
- Dual Insight Q8 GGUF models (HIGH + LOW stages) with split sampling (steps 0–3 HIGH, 3–end LOW)
- Lightning 4-step LoRAs pre-loaded and enabled on both stages (strength 1.0)
- 720×1280 @ 121 frames, h264 output
- RIFE interpolation with in-app on/off toggle
- Upscale pass via 4x-ClearRealityV1
- Load image input with camera/library picker in the app
- Power Lora Loader on both stages with trigger keyword tracking
Required downloads:
| File | Goes in |
|---|---|
Wan2_2-I2V-A14B-HIGH_Insight-Q8_0.gguf |
models/unet/ |
Wan2_2-I2V-A14B-LOW_Insight-Q8_0.gguf |
models/unet/ |
Wan2.2-Lightning_I2V-A14B-4steps-lora_HIGH_fp16.safetensors |
models/loras/ |
Wan2.2-Lightning_I2V-A14B-4steps-lora_LOW_fp16.safetensors |
models/loras/ |
umt5_xxl_fp8_e4m3fn_scaled.safetensors |
models/clip/ |
wan_2.1_vae.safetensors |
models/vae/ |
4x-ClearRealityV1.pth |
models/upscale_models/ |
rife47.pth |
auto-downloaded by RIFE VFI node |
Custom nodes required: ComfyUI-GGUF, VHS, Frame Interpolation (RIFE), rgthree-comfy
All workflows automatically get a virtual Power Lora Loader if one isn't already present — you can add LoRAs to any workflow regardless. Unrecognized custom nodes fall into the collapsible Advanced section with full parameter access.