Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-Drama3 > Drama 3 Speed Options

ComfyUI Node: Drama 3 Speed Options

Class Name

Drama3SpeedOptions

Category
audio/drama3
Author
PedroMarinhoDev (Account age: 50days)
Extension
ComfyUI-Drama3
Latest Updated
2026-11-12
Github Stars
0.17K

How to Install ComfyUI-Drama3

Install this extension via the ComfyUI Manager by searching for ComfyUI-Drama3
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-Drama3 in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

Drama 3 Speed Options Description

Balances Drama 3 speech generation speed, take consistency, and seed repeatability.

Drama 3 Speed Options:

Drama3SpeedOptions lets you choose how Drama 3 balances speech-generation speed, consistency between takes, and repeatability when reusing a seed. It controls optional speed-ups that can change the speech a particular seed produces, while optimizations that preserve the generated takes are already enabled automatically. The available approaches include processing multiple script lines side by side and, under specific hardware and model conditions, allowing the CPU to help with calculations normally handled by the GPU. These approaches can reduce generation time, but they can also change which earlier takes influence a line or make repeated generations differ slightly. Place this node between Load Drama 3 and Drama 3 Generate when you want explicit control over these trade-offs. The node configures generation behavior; it does not generate speech or decode audio itself.

Drama 3 Speed Options Input Parameters:

model

A required input of type DRAMA3_MODEL, normally connected to Load Drama 3. This supplies the loaded Drama 3 model configuration to which the speed and repeatability settings will be applied. The node makes a shallow copy of the supplied model object and sets the selected options on that copy, rather than assigning these settings directly to the incoming object. A shallow copy is not a second independent copy of all model weights, so you should not interpret this operation as loading another complete model into memory. There is no default value, numerical range, or selectable option list for this input; you must supply a compatible Drama 3 model connection.

lines_at_once

A required integer that controls how many script lines can be processed side by side when speakers are configured for one take per line. The default is 1, the minimum is 1, and the maximum is 8. At 1, lines are processed individually; higher values allow multiple lines to progress together, and a line that finishes frees its place for the next line. This can improve throughput, but it changes the context available to each line: a line uses its speaker’s takes that had already finished when it started, which may not include the immediately preceding line. Changing this setting also changes the takes produced by a given seed. Concurrent processing requires a single-GPU setup and a ComfyUI version newer than v0.41.2; otherwise, lines run individually in groups of the selected size and use the takes available before their group began. The supplied benchmark for a ten-minute story on an RTX 4090 reports W4A8 generation times of 225 seconds at 1, 111 seconds at 4, and 107 seconds at 8; FP8 times were 303, 199, and 229 seconds respectively. These measurements illustrate potential gains, not guaranteed performance on your setup.

repeatable_takes

A required Boolean with options False and True, defaulting to False. This controls a specific speed-versus-repeatability trade-off when using int8 experts and an expert cache that holds less than half of them, a situation associated in the tooltip with graphics cards below approximately 20 GB of memory. Experts are specialized parts of the model, and the cache keeps some of them available on the GPU. With False, the CPU can compute part of the work for experts missing from that cache, improving speed under these conditions. However, CPU and GPU arithmetic differ in their last numerical bits, and the CPU’s share of the work depends on timing, so repeating the same seed may not reproduce the same take. With True, the node selects the slower, repeatable behavior intended to produce the same take from the same seed when the other generation settings remain unchanged. The supplied example reports 34 milliseconds per frame with CPU assistance versus 53 milliseconds per frame without it, using a 4 GB expert cache on an RTX 4090. This specific performance benefit is conditional; the source does not establish that it applies to every model format or cache configuration.

Drama 3 Speed Options Output Parameters:

DRAMA3_MODEL

The output is a DRAMA3_MODEL object carrying the supplied model configuration with the selected speed and repeatability settings applied. Connect it to the model input of Drama 3 Generate so that subsequent speech generation uses those choices. Internally, the node returns a shallow copy of the incoming model object with its line-concurrency and repeatability attributes updated. This output is neither an audio waveform nor generated speech codes, and the node does not perform inference while producing it. Its importance is that it passes your chosen generation behavior downstream without requiring a separate model-loading step.

Drama 3 Speed Options Usage Tips:

  • Connect Load Drama 3 to this node, then connect this node’s DRAMA3_MODEL output to Drama 3 Generate. If Generate still receives the loader’s direct output, these settings will not reach that generation path.

  • For repeatable comparisons, set repeatable_takes to True, keep lines_at_once unchanged, and hold the seed, script, voice configuration, and other generation settings constant. Changing line concurrency can change the takes even when the seed stays the same.

  • For long scripts using one take per line, try lines_at_once at 4 before testing 8. In the supplied benchmark, 4 delivered most of the W4A8 speed improvement and was faster than 8 with FP8, so the maximum value is not automatically the best choice.

  • Use lines_at_once at 1 when you want to avoid concurrency-related gaps in the available speaker history. This is useful when you want each successive line to start after the preceding take has completed.

  • Leave repeatable_takes at False when faster iteration matters more than reproducing an identical take, particularly with int8 experts and a small expert cache. Turn it on before making controlled comparisons or trying to reproduce a selected result.

  • Check your ComfyUI version and GPU setup before expecting parallel-line acceleration. The documented concurrent path requires one GPU and a version newer than v0.41.2; other setups fall back to processing lines individually in groups.

  • Compare both generation time and the resulting performance on a representative section of your script. Higher line concurrency changes the speaker history available at line start, so assess continuity as well as speed.

Drama 3 Speed Options Common Errors and Solutions:

Missing or incompatible model input

  • Explanation: The model input is required and must receive DRAMA3_MODEL. Leaving it disconnected or supplying another model type can prevent workflow validation or execution. The provided source does not specify an exact error message for this situation.
  • Solution: Connect the model output of Load Drama 3 to this node’s model input, then connect this node’s output to Drama 3 Generate.

lines_at_once is outside the supported range

  • Explanation: The interface accepts integers from 1 through 8. A manually edited workflow or API request containing a value outside that range does not match the declared input constraints. The execution method does not include a separate custom error message for this case.
  • Solution: Set lines_at_once to a whole number between 1 and 8. Start with 1 for sequential processing or 4 to test higher throughput.

The same seed produces different takes

  • Explanation: This can be expected behavior rather than an error. With repeatable_takes set to False, eligible int8 configurations can split work between CPU and GPU in a timing-dependent way. Changing lines_at_once also changes what a seed generates.
  • Solution: Set repeatable_takes to True, restore the same lines_at_once value, and keep the seed and other generation conditions unchanged before comparing takes.

Increasing lines_at_once does not make generation faster

  • Explanation: Concurrent line processing applies to the one-take-per-line speaker configuration and requires a single GPU with ComfyUI newer than v0.41.2. Otherwise, the documented fallback processes lines individually in groups. Even with concurrency available, higher values do not guarantee faster execution.
  • Solution: Verify the speaker configuration, ComfyUI version, and GPU setup. Time the same representative script at 1, 4, and 8, then keep the fastest setting that gives you acceptable results.

A line does not reflect the immediately preceding take

  • Explanation: With multiple lines in progress, a line starts with the speaker’s takes that have already finished. The preceding line may still be running, so its take is unavailable as context. In the fallback path, lines use the history available before their group began.
  • Solution: Set lines_at_once to 1 when sequential speaker history is more important than processing several lines together.

Enabling repeatable_takes makes generation slower

  • Explanation: This is an intended trade-off under the documented int8 and small-cache conditions. Repeatable behavior forgoes the timing-dependent CPU assistance that can speed up expert calculations.
  • Solution: Keep repeatable_takes enabled when reproducibility matters. Disable it for faster exploratory generation if you can accept differences between runs with the same seed.

Changing the node settings has no effect on generation

  • Explanation: The settings are applied to the model object returned by this node. They will not affect Drama 3 Generate if it receives a different model connection. Some effects are also conditional: line concurrency requires the appropriate speaker configuration, and CPU assistance requires the specified int8 and cache conditions.
  • Solution: Confirm that Drama 3 Generate receives this node’s output, not a direct connection from Load Drama 3. Then check whether your speaker, model, and cache configuration meets the conditions for the option you are testing.

Drama 3 Speed Options Related Nodes

Go back to the extension to check out more related nodes.
ComfyUI-Drama3
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Drama 3 Speed Options