IAMCCS Influencer Sampler
RES4LYF-flavored sampling without the RES4LYF dependency
- model
- positive
- negative
- latent_image
- sigmas
- guides
- options
- output
- denoised
- options
If you've been anywhere near flow-matching workflows in the last year you've seen the RES4LYF sampler names - exponential/ddim, exponential/res_2s, the beta57 schedule - and you've probably also seen the license rider that makes hosting services twitch. This node is IAMCCS's answer: a clean-room sampling bridge that exposes the same profiles the Influencer apps use, but built entirely on ComfyUI's public sampling API. The module header says it in so many words: it does not import, vendor, or execute RES4LYF code. Same flavor, no AGPL-with-a-commercial-rider dependency.
The hook is the four sampler profiles, sampler_name: linear/euler (the default, for the first creative pass), exponential/ddim and exponential/res_2s (the edit/reference workflows), and exponential/res_4s_munthe-kaas - the app's high-order refinement profile, implemented internally via ComfyUI's maintained res_multistep_ancestral. The docstring is admirably honest that it does not claim bit-identical output to the third-party implementation - it's the same solver family, not the same code. On the scheduler side you get the full modern list including beta57 and bong_tangent, both RES4LYF-borrowed names that flow-matching users expect to see.
Now the inputs that matter:
steps(15 default) andsteps_to_run(-1 = run all).steps_to_runis the underrated one: set it to a small number to cheaply preview the first few steps of a profile before committing.denoise(1.0) andcfg(1.0) - this is a flow-matching-era sampler, so the defaults are honest. cfg 1.0 is normal here.eta(0.5) - the tooltip calls it "ancestral noise strength for the refinement profile." Only the ancestral/res_4s route really hears it.seedwithcontrol_after_generate- standard.sampler_modehas exactly one choice (standard) - a compatibility seam, not a feature. Ignore it.bongmath(default true) - a BOOLEAN whose tooltip says "Compatibility field retained for saved IAMCCS settings." It does literally nothing in the code. Leave it on; it exists so old saved workflows load without JSON surgery.
Optional inputs are model, positive, negative, latent_image, sigmas, guides, options - the standard sampler contract, and it validates that model/positive/negative/latent are all present before running. Outputs: output (the sampled latent), denoised (the last clean x0, useful for previews), and options (an OPTIONS dict that records which profile/scheduler/steps actually ran - handy for provenance when you're comparing profiles).
Installation is pack-standard (search IAMCCS in ComfyUI Manager, or clone into custom_nodes, restart). It has no dependencies beyond ComfyUI itself - no RES4LYF install needed, which is the entire point.
The honest takes: if you already run RES4LYF and love its 69-solver menu, this node isn't trying to replace it - it's a four-profile compatibility surface for the Influencer apps. And exponential/res_4s_munthe-kaas being mapped to res_multistep_ancestral means "same spirit, different implementation" - expect visually equivalent results, not pixel-identical ones. For anyone who just wants the beta57 schedule in a no-fuss node, though, this gets you there with none of the licensing questions.
Inputs (17)
| Name | Type | Default | Description |
|---|---|---|---|
| eta | FLOAT | 0.500–10 | Ancestral noise strength for the refinement profile. |
| sampler_name | COMBO | linear/euler | 4 options: linear/euler, exponential/ddim, exponential/res_2s, exponential/res_4s_munthe-kaas |
| scheduler | COMBO | beta | 11 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +5 |
| steps | INT | 151–10000 | — |
| steps_to_run | INT | -1-1–10000 | — |
| denoise | FLOAT | 1.000–1 | — |
| cfg | FLOAT | 1.000–100 | — |
| seed | INT | 00–18446744073709550000 | — |
| sampler_mode | COMBO | standard | 1 options: standard |
| bongmath | BOOLEAN | true | Compatibility field retained for saved IAMCCS settings. |
| modelopt | MODEL | — | |
| positiveopt | CONDITIONING | — | |
| negativeopt | CONDITIONING | — | |
| latent_imageopt | LATENT | — | |
| sigmasopt | SIGMAS | — | |
| guidesopt | GUIDES | — | |
| optionsopt | OPTIONS | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| output | LATENT | — |
| denoised | LATENT | — |
| options | OPTIONS | — |