ComfyUI Node
RS Progressive Resolution Sampler
Progressive T2I: trades some detail for faster large-image sampling by starting smaller and expanding with Lanczos. Lower start scales can lose more detail or change composition; quality is not guaranteed. Existing multi-character prompts, masks, text, per-character LoRAs and outputs remain available. 1.0 uses the ordinary sampler exactly. Solver history restarts at expansion; quality and speed depend on the model/sampler.
RS Progressive Resolution Sampler
- model
- clip
- vae
- character_pipe
- image
- latent
- character_pipe
- generation_info
◄promptmasterpiece, best quality, 5girls, university campus►
◄negative_promptlow quality, bad anatomy, text, watermark►
◄width1536►
◄height768►
◄start_scale0.67►
◄seed0►
◄steps30►
◄cfg4.0►
◄sampler_nameeuler_ancestral►
◄schedulersgm_uniform►
◄autosavetrue►
◄save_formatPNG►
◄webp_quality90►
◄prompt_syntaxComfyUI►
◄use_image_settingsfalse►
◄region_feather0.04►
◄initial_common_prior0.05►
◄common_after_step3_boost0.10►
◄initial_character_gain1.7►
◄initial_spatial_separation0.90►
◄boundary_hold_steps3►
◄boundary_decrement0.20►
◄layout_methodregional►
◄flux_guidance3.5►
CategoryRS MultiCharacter
Inputs (28)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| clip | CLIP | — | |
| vae | VAE | — | |
| character_pipe | RS_MC_CHARACTER_PIPE,NAI_CONTEXT | — | |
| prompt | STRING | masterpiece, best quality, 5girls, university campus | — |
| negative_prompt | STRING | low quality, bad anatomy, text, watermark | — |
| width | INT | 153664–4096 | — |
| height | INT | 76864–4096 | — |
| start_scale | FLOAT | 0.670.5–1 | Starting width/height relative to the final size. Trade some detail for faster large-image sampling. Lower scales can lose more detail or change composition; quality is not guaranteed. 1.0 uses ordinary sampling. Existing multi-character prompts, masks and LoRAs remain available. |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 302–100 | Anima/SDXL: 30; Krea2 Turbo: 8; MOSS: 50. Common weight rises every 3 scheduled steps. Boundary hold uses sampling steps and is capped by the selected step count. |
| cfg | FLOAT | 4.01–20 | — |
| sampler_name | COMBO | euler_ancestral | All installed samplers are selectable. Anima starting points: euler_ancestral + sgm_uniform; euler + linear_quadratic; dpmpp_2m_sde_gpu + beta. Compared at 30 steps / CFG 4, one two-character scene and seed; quality varies. SDXL: euler_ancestral + normal. Krea2 Turbo / MOSS: euler + simple, with the matching example settings. |
| scheduler | COMBO | sgm_uniform | All installed schedulers are selectable. Anima: try sgm_uniform with euler_ancestral, linear_quadratic with euler, or beta with dpmpp_2m_sde_gpu. SDXL: normal with euler_ancestral. Krea2 Turbo / MOSS: simple with euler. Choose the pair manually; changing a sampler does not reset this value. |
| autosave | BOOLEAN | true | — |
| save_format | COMBO | PNG | 3 options: PNG, WebP lossless, WebP lossy |
| webp_quality | INT | 901–100 | — |
| prompt_syntaxopt | COMBO | ComfyUI | 2 options: ComfyUI, NovelAI |
| use_image_settingsopt | BOOLEAN | false | An image pipe always supplies its prompts/characters. Enable to also restore dimensions, seed, sampling and routing settings. Connected model/LoRA/CLIP/VAE remain your choice; check the image loader report. |
| region_featheropt | FLOAT | 0.040–0.25 | Softens point partitions only. Boxes use fractional cell coverage and ignore this setting. |
| initial_common_prioropt | FLOAT | 0.050–1 | — |
| common_after_step3_boostopt | FLOAT | 0.100–1 | — |
| initial_character_gainopt | FLOAT | 1.71–3 | — |
| initial_spatial_separationopt | FLOAT | 0.900–0.99 | — |
| boundary_hold_stepsopt | INT | 31–30 | — |
| boundary_decrementopt | FLOAT | 0.200.01–1 | — |
| layout_methodopt | COMBO | regional | Regional routes attention for all supported models. MOSS native sends one official layout JSON prompt, requires MOSS and box or mask regions (masks become enclosing rectangles), and ignores regional strength/boundary/feather settings. Experimental: exact object bounds are not guaranteed. |
| flux_guidanceopt | FLOAT | 3.50–20 | Embedded guidance for FLUX.1 Dev / FLUX.2 Dev; separate from CFG. Schnell, Klein and FLUX 3 ignore this value. FLUX 3 Base uses ordinary CFG. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| latent | LATENT | — |
| character_pipe | RS_MC_CHARACTER_PIPE | — |
| generation_info | STRING | — |