Nodes/ComfyUI-Angelo/Angelo — Overrides
ComfyUI Node

Angelo — Overrides

Drive Angelo's sampler settings from your graph, or generate with a second model

By shootthesound·Created 3 months ago·Updated about a month ago· 128
Angelo — Overrides
  • guider
  • sampler
  • sigmas
  • gen_model
  • gen_positive
  • gen_negative
  • gen_vae
  • overrides
steps-1
cfg-1.0
sampler_name(toolbar)
scheduler(toolbar)
disable_live_previewfalse
gen_steps25
gen_cfg5.0
gen_sampler_nameeuler
gen_schedulernormal

This one doesn't do anything on its own. It's a settings bundle for its sibling node, AngeloRefine - the click-to-edit sampler. You wire AngeloOverrides' overrides output into AngeloRefine's overrides input, and it lets you do two things you can't do from Angelo's own toolbar: drive steps/cfg/sampler/scheduler from elsewhere in the workflow instead of re-typing them, and hand Sampler Mode a completely different model to generate the base image with before your edit model takes over for everything after.

The single-source-of-truth case is the simple one. Angelo's toolbar has its own Steps/CFG/Sampler/Scheduler widgets, which is fine if Angelo is the only sampler in your graph - but if you're already setting those values somewhere else and don't want two places to keep in sync, wire this node in instead. Each field opts in independently: steps and cfg default to -1, sampler_name and scheduler default to the string (toolbar), and any field left at its sentinel just falls through to whatever's set on Angelo's own toolbar. So you can override only cfg and leave the rest alone if that's all you need.

The gen bundle is the more interesting reason to reach for this node. Klein 9B and Qwen-Image-Edit are excellent editors, but they're not always the best-looking generators - and rather than force you to pick one model for both jobs, the gen bundle lets you generate the base with something else entirely (FLUX Krea 2, SDXL, whatever produces the look you want) and then edit with Klein or Qwen-Image-Edit for the rest of the session. Wire gen_model, gen_positive, and gen_vae (plus optionally gen_negative), and when AngeloRefine is in Sampler Mode it generates the base with that model - using this node's own gen_steps (default 25), gen_cfg (default 5), gen_sampler_name (default euler), and gen_scheduler (default normal) - decodes it, and re-encodes the pixels with your main edit VAE so every click-to-edit afterward runs on the edit model exactly as if you'd loaded a photo. The wiring has real teeth: gen_positive has to be encoded with the gen model's own CLIP, since conditioning doesn't carry across model families - you'll end up with two CLIP Text Encode chains in the graph, one for each model. gen_negative is optional; leave it unwired and Angelo zeroes a copy of gen_positive for you, same as ConditioningZeroOut. And the wired latent needs to be sized for the gen model, not the edit model - Angelo prints the resolution to console each run, which is the easiest way to catch a mis-sized latent before it wastes a generation. gen_model / gen_positive / gen_vae are all-or-nothing: wire two of the three and you get a clear error instead of a silent fallback to the wrong model.

There's a third, more specialist slot: guider, sampler, and sigmas - optional GUIDER / SAMPLER / SIGMAS inputs for swapping out Angelo's entire sampling stack rather than just naming pieces of it. This is where a third-party sampler pack plugs in - something like a custom guider or a power-sigma schedule from a pack such as RES4LYF, which exists specifically because flow-matching architectures (Klein among them) want different sampling math than the old SD1.5/SDXL Karras-schedule defaults. Wire all three and Angelo runs the sampling loop through your guider instead of its own - the toolbar's steps/cfg/sampler/scheduler stop mattering, though the Denoise slider on AngeloRefine still applies. Wire only one or two of the three and it silently falls back to Angelo's default sampling - a genuinely different failure mode from the gen bundle's hard error, so don't assume a partial wire here will warn you.

Rounding out the required fields: disable_live_preview, a plain boolean, off by default. This exists because ComfyUI's global latent-preview setting (Settings → Preview method) can render straight into Angelo's node mid-sample and squash the editing canvas down to a sliver - a real complaint from people running Angelo alongside other samplers that still want their previews. Flip it on and it suppresses the preview callback for this Angelo only, leaving your other KSamplers untouched.

Installing it takes no separate steps - AngeloOverrides ships in the same pack as AngeloRefine, so cd ComfyUI/custom_nodes && git clone https://github.com/shootthesound/ComfyUI-Angelo and a restart gets you both, filed under the same sampling/Angelo category in the node menu. No extra dependencies.

The main way to misuse this node is to add it to your graph and expect something to happen - it's inert until its overrides output actually reaches AngeloRefine's overrides input. Past that, the two failure shapes worth remembering are the ones above: the gen bundle refuses to run on a partial wire (loud, in Sampler Mode), while the guider/sampler/sigmas trio just quietly reverts to normal sampling if you don't wire all three (silent - check your console if a custom sampler doesn't seem to be taking effect).

Categorysampling/Angelo

Inputs (16)

NameTypeDefaultDescription
stepsINT-1-1–100-1 = use Angelo's toolbar value.
cfgFLOAT-1.0-1–30-1 = use Angelo's toolbar value.
sampler_nameCOMBO(toolbar)(toolbar) = use Angelo's toolbar value.
schedulerCOMBO(toolbar)(toolbar) = use Angelo's toolbar value.
disable_live_previewBOOLEANfalseWhen ON, suppress ComfyUI's live latent preview for this Angelo node only (#21). Useful if the global preview is squashing Angelo's editor area but you still want previews on your other samplers.
guideroptGUIDER
sampleroptSAMPLER
sigmasoptSIGMAS
gen_modeloptMODELGeneration model for Sampler Mode. When wired (with gen_positive + gen_vae), Angelo generates the base with THIS model, then hands the pixels to the edit model for the editing session. Size Angelo's wired latent for THIS model's VAE ratio (a plain EmptyLatentImage for SD-class models).
gen_positiveoptCONDITIONINGPositive conditioning for gen_model — encode with the GEN model's CLIP, not the edit model's.
gen_negativeoptCONDITIONINGOptional negative for gen_model. Unwired = zeroed gen_positive (like ConditioningZeroOut).
gen_vaeoptVAEVAE matching gen_model — used to decode the generated base before it is re-encoded with the edit VAE.
gen_stepsoptINT251–100[Gen bundle] Steps for the base generation with gen_model.
gen_cfgoptFLOAT5.00–30[Gen bundle] CFG for the base generation with gen_model.
gen_sampler_nameoptCOMBOeuler[Gen bundle] Sampler for the base generation with gen_model.
gen_scheduleroptCOMBOnormal[Gen bundle] Scheduler for the base generation with gen_model.

Outputs (1)

NameTypeDescription
overridesANGELO_OVERRIDES