🔶 Load Checkpoint +Frame CKPT_PROMPT
Package a checkpoint, prompt, and start frame together
- lora
- CKPT_PROMPT
This node doesn't output a model or a conditioning at all - and that trips people up the first time they use it. What it outputs is a single bundle: a CKPT_PROMPT, which packages a checkpoint, its positive/negative prompt, and the frame number it should activate at, all into one typed wire. It's a building block, not an endpoint - you're meant to make several of these (one per checkpoint-and-prompt change you want across a sequence) and feed all of them into CheckpointPrompt_FrameMixer, which is the node that actually resolves them against the current frame and produces usable MODEL/POSITIV/NEGATIV/VAE output.
This is the more flexible sibling of chaosaiart_CheckpointLoader. That node shares one prompt across every checkpoint swap; this one lets each checkpoint carry its own prompt, at the cost of needing an extra Mixer node downstream to actually use it. If you only need to change the model and keep the same prompt throughout, the plain "by Frame" loader is less setup. If a checkpoint change should also mean a prompt change - switching from a realism-leaning checkpoint with one style of prompt to an anime one with a totally different vocabulary, say - this pairing is what handles it cleanly.
Inputs and outputs
Start_Frame- required INT (default 1, min 1, max 1,000,000,000). The frame count at which this bundle should become active in the Mixer.Checkpoint- required, your model dropdown.Positiv,Negativ- required, multiline text specific to this checkpoint.lora(optional,LORAtype) - attach a LoRA to this particular checkpoint stage.CKPT_PROMPT- the single output. It only connects to other Chaosaiart nodes that accept that type; you can't wire it into a standard ComfyUI checkpoint or conditioning slot.
Installing it
ComfyUI Manager: Install via Git URL → https://github.com/chaosaiart/Chaosaiart-Nodes, which is the method the README itself walks through. A title search for "Chaosaiart" should also find it.
Manual:
cd ComfyUI/custom_nodes
git clone https://github.com/chaosaiart/Chaosaiart-Nodes
Linux (activate your venv first, if applicable):
pip install opencv-python
pip install tqdm
Windows: use the Install_windows script in the cloned folder instead. Restart ComfyUI. No extra downloads - it points at checkpoints and LoRAs you've already got installed.
Common issues
"Nothing happens" is normal - this node alone does nothing. If you wire it up expecting a MODEL or image output and get neither, that's by design: connect its CKPT_PROMPT output into chaosaiart_CheckpointPrompt_FrameMixer to actually get usable outputs. This node on its own just builds a data bundle.
Frame numbers only matter relative to what you feed the Mixer. Start_Frame isn't compared against a video timeline, it's compared against the activ_frame counter the Mixer receives - which counts queue runs, not seconds or timeline frames.
Whole-pack failures on a broken dependency. Every node here, this one included, shares one Python module. If opencv-python or tqdm didn't install, the whole pack disappears from ComfyUI's node list - check the startup console for an import error before digging further.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| Start_Frame | INT | 11–1000000000 | — |
| Checkpoint | COMBO | 0 options: | |
| Positiv | STRING | — | |
| Negativ | STRING | — | |
| loraopt | LORA | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CKPT_PROMPT | CKPT_PROMPT | — |