For ComfyUI’s ControlNetApplySD3 node, setting strength to 0.0 skips that node’s ControlNet application and passes the input conditioning through unchanged to both outputs, according to its documentation. You can use that as a control comparison against a nonzero-strength run, provided both runs receive the same positive and negative conditioning and their outputs follow the same downstream connections. This contract is specific to ControlNetApplySD3; the node is marked deprecated, so check the current node and its documentation before relying on the same behavior elsewhere.
Keep the comparison about one node
The node accepts positive and negative conditioning, a ControlNet model, a VAE and an image, along with strength, start_percent and end_percent. Its documentation specifies that zero strength means no ControlNet guidance is applied and the input conditioning is returned unchanged to both outputs. That is a statement about the conditioning at this node’s outputs, not proof that the complete workflow has no other controls or that its final image must match a run with the node removed.
Also note the documentation’s sequencing detail: if reused conditioning already contains control information, this node’s ControlNet is linked after the previous one. A zero-strength comparison therefore needs a clearly identified input; do not assume the incoming conditioning is plain text conditioning.
Set up a paired comparison
- Find the exact node in the graph and confirm its name is
ControlNetApplySD3. Read the deprecation notice and check whether your installed ComfyUI version still offers this node. If you use a replacement, first establish that replacement’s own documented contract. - Record the positive and negative conditioning sources, ControlNet model, VAE, guidance image,
start_percentandend_percent. Keep the same inputs for both runs. In particular, use the same conditioning objects or the same upstream paths, not separately rebuilt prompts that may differ. - Run the control with
strengthset to0.0. Keep both positive and negative outputs connected to the same downstream inputs they will use in the comparison. - Run the comparison with a chosen nonzero
strength, changing no other setting or connection. The documentation lists a range of 0.0 to 10.0 and a default of 1.0; choose a value for your own comparison rather than treating the default as a guaranteed best setting. - Record the node values, input paths, output connections and observed result. Label image differences as observations, not proof that every difference was caused only by ControlNet if any other workflow input changed.
Check the wiring before judging images
Trace the positive output to the same downstream positive input in both runs, and do the same for negative. A crossed or bypassed connection changes what the comparison means. Keep the sampler and other downstream workflow settings fixed if you are comparing generated images. The source documentation defines the node’s conditioning behavior; it does not promise bit-for-bit identical images between separate generations.
Here is a hypothetical worksheet, not a reported test. Suppose both runs use the same upstream positive and negative paths, guidance image, model, VAE and timing values, with identical downstream wiring. The 0.8 result looks more aligned with the guide image. That observation is worth recording, but it does not establish a universal effect size or prove that 0.8 is optimal.
| Check | Hypothetical control run | Hypothetical comparison run | Acceptance question |
|---|---|---|---|
| Node | ControlNetApplySD3; deprecated notice recorded | Same node and version | Did the node contract stay the same? |
| Strength | 0.0 | 0.8 | Was strength the only changed setting? |
| Conditioning and wiring | Same positive/negative sources and downstream inputs | Same positive/negative sources and downstream inputs | Are both outputs routed consistently? |
| Observation | Reference for this paired comparison | Appears more guide-aligned | Is this an observation, not a general claim? |
If the comparison is unclear
If the two runs differ in prompt path, image, timing, downstream wiring or other settings, correct that mismatch and repeat the pair. If you cannot identify the conditioning entering this node, trace its upstream source first; it may already include earlier control information. If the node is unavailable or replaced, do not transfer the zero-strength conclusion automatically. Locate the replacement’s official documentation and verify its zero-value behavior before using it as a pass-through control.
The node documentation is marked AI-generated and invites corrections. Treat its explicit zero-strength statement as the documented behavior for this node, while remembering both that caveat and the deprecation warning. It does not establish how every ControlNet node, custom node, or workflow treats zero.
Common questions
Does zero strength remove the node from the graph? No. The documented behavior is that this node skips applying its ControlNet and returns the conditioning unchanged. The node remains in the graph and its outputs still need to be connected as intended.
Does zero strength clear control information already in the input? The documentation says the input conditioning is passed through unchanged. It does not say that existing information is removed. If upstream conditioning may already contain control information, account for that when interpreting the comparison.
Can I use this rule with another ControlNet node? Not on this evidence alone. Confirm the exact node and check its own current documentation, especially because `ControlNetApplySD3` is deprecated.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.