pipeKSamplerAdvanced v1 (Legacy)

ComfyUI's KSamplerAdvanced exists because sometimes you don't want to sample all the steps at once - you want to run step 0–15, upscale the latent, then run step 15–20 on top for a hires pass. The tinyterra equivalent for its pipe system was pipeKSamplerAdvanced v1 (ttN pipeKSamplerAdvanced): a legacy sampler that takes a ttN PIPE_LINE, gives you start_at_step / end_at_step control, and returns both a new pipe and every raw slot for downstream use.

It's in the 🌏 tinyterra/legacy category, and that label is doing real work here. The v1 advanced sampler was replaced by pipeKSamplerAdvanced_v2, which kept the start/end-step engine but embedded the Advanced CLIP text encoding (prompt boxes right on the node) and swapped in the richer upscale options. If a workflow you downloaded has a v1 in it, it still loads and runs - that's why legacy nodes aren't deleted - but for new graphs you want the v2.

How it works

Feed it a pipe, and its outputs are pipe, model, positive, negative, latent, vae, clip, image and seed. The optional_* inputs (optional_model, optional_positive, optional_negative, optional_latent, optional_vae, optional_clip) are pipe overrides - connect one and it replaces that slot for this sampling run, exactly like pipeEDIT but scoped to the sampler. It also takes a legacy xyPlot input for grid runs.

The controls that matter:

• start_at_step / end_at_step - the window of the denoising schedule to run. The classic pattern: a normal sampler does 0→20; a hires pass does 16→20 (or wherever) with add_noise disabled and the upscaled latent in.
• sampler_state - "Sample" or "Hold". Hold stops this node from actually sampling, which is how you keep a shared sampler template in the graph without it running every time.
• add_noise - enable/disable. Disable it for the second stage of an img2img/hires pass so you don't re-roll the noise.
• lora_name + lora_model_strength / lora_clip_strength - note the v1 split: model and CLIP strengths are separate floats here (the v2 collapses them into one lora_strength).
• upscale_method, factor, crop - a simple latent upscale before sampling (nearest/bilinear/area/bicubic/lanczos/bislerp, or None).
• start_at_step-adjacent: return_with_leftover_noise - enable when a later pass will continue from this latent.
• image_output / save_prefix - preview or save, same as the rest of the pack.

The one thing it does not have is a prompt box - v1 expects the conditioning already in the pipe. That's the single biggest reason to prefer v2.

Install and gotchas

One-time pack install: ComfyUI Manager → search ComfyUI_tinyterraNodes, or git clone https://github.com/TinyTerra/ComfyUI_tinyterraNodes.git into ComfyUI/custom_nodes, restart. No extra dependencies or downloads.

Gotchas worth knowing: the legacy nodes are SD1.5/SDXL-era, so expect friction on newer architectures - that's the pack's known weak spot. And if you find yourself setting start_at_step/end_at_step constantly, you're the person the v2 was built for; migrating is just swapping the node in and moving your prompt into the box. For old saved workflows, leave it alone and let it run - it still does the job.