❌ Double Stage Sampler [Deprecated]
The old two-stage sampler — dead, and rightly replaced
- latent_input
- genparams
- model
- clip
- transcoder
- refiner_model
- refiner_clip
- latent_output
You've hit a ❌ Double Stage Sampler [Deprecated] node, which means one of two things: you're looking at a workflow saved months ago, or someone exported an old graph and shipped it around. This was the node that actually ran TinyBreaker's two-pass denoising - base model first, refiner second, transcoder in between - before the pack was restructured. It's been superseded, the README and the pack's own code both point you elsewhere, and you should treat it the way you'd treat a KSampler from a beta UI: port it forward, don't build on it.
What it did
It took the full set of components - latent_input, genparams (which carried the sampler config), model + clip for the base stage, transcoder to bridge latent spaces, and refiner_model + refiner_clip for the second stage - and denoised in sequence, producing a single latent_output. In that sense it was the conceptual ancestor of the current 💪TB | Tiny Dual Sampler: same hybrid idea, pre-GenParams-rewrite plumbing. You can see the lineage: even the current sampler still accepts a transcoder input, which is a concept this node introduced.
Why it's deprecated
The pack moved to a cleaner split. The modern design puts the sampler configuration into the genparams blob (fed by nodes like Set Image Details and Select Style) and separates the sampler from the "gather all the models" job that the loaders now do. The old node mixed those concerns and expected you to hand-wire model, clip, refiner, and transcoder all into one fat input list - exactly the fragile kind of graph that broke on every checkpoint format change. Keeping it around under a __deprecated category lets old workflows still load, but the author's intent is unambiguous: the replacement is the Tiny Dual Sampler, and it does the same two-stage job against the current loader outputs.
If you're stuck with one
Don't fight it. In the canvas, right-click the node and find the "replace" entry for the modern sampler, or rebuild using the pack's reference workflow (drag the workflow image into ComfyUI - it's in the README). If it still exists in your saved workflow and ComfyUI refuses to run it cleanly on a new checkpoint, that's expected behavior, not a bug: old components and new checkpoints weren't designed to meet. The good news is porting is small - the modern sampler wants the same conceptual inputs (latent, genparams, model, clip, transcoder, refiner model/clip), they're just wired from the loader outputs instead of one mega-node.
Install
It ships in the pack, so it arrives with any install - ComfyUI Manager search "tinybreaker" or git clone https://github.com/martin-rizzo/ComfyUI-TinyBreaker into custom_nodes, then restart. You won't find it in the normal node menu by design (deprecated nodes are hidden from default searches), so if you want it you're going out of your way. Don't. Use the Tiny Dual Sampler.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| latent_input | LATENT | The latent image to denoise. | |
| genparams | GENPARAMS | The generation parameters containing the sampler configuration. | |
| model | MODEL | The model used for denoising the latent images. | |
| clip | CLIP | The T5 encoder used for embedding the prompts. | |
| transcoder | TRANSCODER | The transcoder model used for converting latent images from base to refiner. | |
| refiner_model | MODEL | The model used for refining latent images. | |
| refiner_clip | CLIP | The CLIP model used for embedding text prompts during refining. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent_output | LATENT | Latent image after denoising. |