Skip to main content
maxresdefault

Watch tutorial 9 on YouTube

Render anywhere: your GPU, Comfy Cloud, or another model.
Every render is a job, and a job runs on a backend: your own ComfyUI, Comfy Cloud, or Comfy Router (hosted models such as Seedance, Kling and Veo). One episode can use all three, a shot at a time.

1. Your GPU

1

Queue some shots

With ComfyUI running on this PC (or set in Settings → ComfyUI), Generate missing or Render this shot queues on it. Nothing else to set.
2

Watch Jobs

Jobs (top right) has one lane per backend. Your own ComfyUI runs one job at a time: the pictures first, then the takes. Each job shows its progress and its log; Earlier and Later reorder a waiting one, Pause holds the lane.
3

See the graph (optional)

Each shot is a ComfyUI workflow of its own: seriesmith’s Shot Loader, the model, the sampler, the saver. Open it on ComfyUI’s canvas any time. To change how a model renders, see The graph a model renders with.

2. Comfy Cloud

No GPU needed: the same shot, on Comfy’s machines.
1

Set a Comfy API key

Get a key at platform.comfy.org, then Settings → Comfy API key, paste it and save. It turns on Comfy Cloud and Comfy Router. (A clone in custom_nodes may need Comfy’s SDK: Settings shows the command, or an Install button.)
2

Render on Comfy Cloud

In the top bar, Render on → cloud. Every render from this page goes there this session (back to each shot’s own to undo). Or choose it per request in the Render dialog’s own Render on row.
3

Render a shot

Render as usual. Cloud doesn’t have seriesmith’s nodes, so the shot’s graph is rebuilt from Cloud’s own, and its pictures go with it, cut for that shot. It runs on the Cloud lane in Jobs.
4

Compare

The take comes back beside the one from your GPU: Compare with…, and Use the one you want.

3. One shot on another model

Any shot can render on any model, without changing the rest of the episode.
1

Open New take…

Right-click a shot’s take → New take like this one….
2

Pick the model

Its Model class list is grouped by where a model runs: Your GPU first, then Comfy Router (paid). Pick Seedance 2.5. This is a one-off: the shot’s own model is unchanged.
3

Check the size and length

Under it, the size and length are that model’s for this shot (its own frame grid), so what it prints is what you get.
4

Generate

Generate the take. It runs on the Router lane; Jobs says where it is at Router (waiting at Router (#2 in line), running on Router).
5

Pick one for the cut

Seedance’s take sits beside H3’s: compare, and Use the one you want. Only this shot changed.
To move a shot to another model for good, set Video model class in its shot form (under More), or target: in the script. Model: on the episode’s card sets the episode’s.

4. Paid runs

Comfy Cloud and Router bill your Comfy account.
  • The count: Jobs’ title counts this session’s paid runs, with reset.
  • A batch asks first: Generate missing or Re-render stale on a paid backend confirms before queuing.
  • A cap: Settings → Paid runs per session stops queuing past that number (empty is no cap).
  • Hosted models are never picked by themselves: a series renders on Router only where you choose a Router model.

Next

10 · Continuity

Two shots that carry on as one.

Where shots render

Every backend and the settings that choose.