🌀 BV Regional FLUX.2 Klein 9B Attention:
This node applies regional prompt guidance through joint-attention routing for the exact FLUX.2 Klein 9B architecture. It prepares region-aware conditioning and patches the model so different parts of an image can follow their assigned regional instructions during sampling. It also supports regional LoRA assignments and follows the regional document’s selected negative-conditioning policy. Use its results with a standard KSampler; this node prepares the model and conditioning but does not generate an image by itself.
🌀 BV Regional FLUX.2 Klein 9B Attention Input Parameters:
model
A required MODEL connection containing the FLUX.2 Klein 9B model to patch. The node prepares any global LoRA changes associated with the regional setup, then adds its regional attention patch to this model. The node is specifically intended for the exact FLUX.2 Klein 9B architecture; connecting another architecture is not supported by its stated purpose. No default, minimum, or maximum applies to this connection.
clip
A required CLIP connection used to compile the regional prompts into conditioning the sampler can use. If your model and text encoder do not belong to the same compatible setup, prompt compilation may fail or produce unsuitable conditioning. Global LoRA preparation can also affect this connection. No default, minimum, or maximum applies.
regional
A required BV_REGIONAL connection containing the regional document: the instructions that define the areas and prompts to be routed. The node also uses this context when preparing global LoRAs and determining regional LoRA assignments. Its selected negative-mode policy governs how negative conditioning is prepared. No default, minimum, or maximum applies.
attention_strength
A required FLOAT setting controlling the strength passed to the regional attention patch. Use it to adjust how strongly the patch applies regional routing: 0.0 is the lowest selectable value and 1.0 is the highest. The default is 1.0, the minimum is 0.0, the maximum is 1.0, and the interface step is 0.01.
start_percent
A required FLOAT setting passed to the attention patch to specify the beginning of its sampling interval as a fraction of the run. The default is 0.0, the minimum is 0.0, the maximum is 1.0, and the interface step is 0.001. For example, 0.0 selects the beginning of the sampling timeline; it is not a step count.
end_percent
A required FLOAT setting passed to the attention patch to specify the end of its sampling interval as a fraction of the run. The default is 0.5, the minimum is 0.0, the maximum is 1.0, and the interface step is 0.001. For example, 0.5 places the end halfway through the sampling timeline. Keep it later than start_percent when you want a meaningful interval.
lora_registry
An optional BV_LORA_STACK_REGISTRY connection for the legacy LoRA sidecar workflow. If either legacy sidecar input is connected, the node resolves regional LoRA stacks from the sidecar inputs instead of the regional context’s LoRA scopes. Leave it disconnected when using the regional context’s LoRA configuration. No default, minimum, or maximum is specified.
lora_bindings
An optional BV_REGIONAL_LORA_BINDINGS connection for legacy assignments of LoRA stacks to regional scopes. Like lora_registry, connecting it selects the legacy sidecar resolution path when either sidecar input is present. For a legacy setup, provide the registry and bindings required by that setup; otherwise leave both optional connections disconnected. No default, minimum, or maximum is specified.
🌀 BV Regional FLUX.2 Klein 9B Attention Output Parameters:
patched_model
The MODEL produced after global LoRA preparation and application of the FLUX.2 Klein regional attention patch. Connect it to the model input of a standard KSampler together with this node’s conditioning outputs. It is a prepared model connection, not a generated image or a saved model file.
positive
The CONDITIONING compiled from the regional document and text encoder for positive guidance, with applicable regional LoRA hook passes applied. Connect it to the sampler’s positive-conditioning input so the regional prompts can participate in generation. Its value is conditioning data, not readable prompt text or an image preview.
negative
The CONDITIONING compiled according to the regional document’s selected negative-mode policy, with applicable regional LoRA hook passes applied. Connect it to the sampler’s negative-conditioning input. Its contents depend on that policy; do not assume it always represents the same negative-prompt behavior across regional setups.
🌀 BV Regional FLUX.2 Klein 9B Attention Usage Tips:
- Use a model and matching text encoder for the exact FLUX.2 Klein 9B architecture, then connect all three outputs to the corresponding inputs of a standard KSampler.
- Start with
attention_strengthat its default1.0and the default interval of0.0to0.5. Change one setting at a time so you can tell whether a difference comes from routing strength or timing. - If regional guidance appears too strong for your image, try lowering
attention_strengthin small increments. If you want to change when routing is applied, adjuststart_percentandend_percentwhile keeping the start earlier than the end. - Check your regional document’s prompts, areas, and negative-mode selection before troubleshooting the sampler: this node compiles those settings rather than replacing them.
- For regional LoRAs, prefer the configuration carried by the regional context. Connect the optional
lora_registryandlora_bindingsonly when using the legacy sidecar workflow, because either connection changes where the node obtains regional LoRA scopes.
🌀 BV Regional FLUX.2 Klein 9B Attention Common Errors and Solutions:
Required input is missing
- Explanation: The node needs
model,clip,regional,attention_strength,start_percent, andend_percent. A disconnected required socket can prevent execution. This is a possible workflow validation error, not a quoted error message from the node. - Solution: Connect all required sockets and supply values for the three numeric settings.
Input type does not match the socket
- Explanation: The required connections expect
MODEL,CLIP, andBV_REGIONALrespectively; the optional sidecars expect their specified registry and bindings types. An incompatible connection may be rejected by the workflow interface. - Solution: Check the type shown on each socket and connect an output of the matching type.
Model architecture is incompatible with FLUX.2 Klein 9B
- Explanation: The node’s stated target is the exact FLUX.2 Klein 9B architecture. A different model may fail when the architecture-specific attention patch is applied. No exact exception text is specified in the provided source.
- Solution: Use the intended FLUX.2 Klein 9B model and its compatible text encoder, or choose a regional attention node intended for your model architecture.
Regional document cannot be compiled
- Explanation: The node reads the regional context and compiles it into conditioning and routing information. An incomplete or incompatible regional setup may stop that process; the provided source does not specify an exact error message.
- Solution: Inspect the upstream regional setup, confirm its areas and prompts are configured, and test again with a simpler regional document if needed.
Regional LoRA stack cannot be resolved
- Explanation: Regional LoRA stacks are resolved before conditioning and the attention patch are returned. Incorrect legacy registry or binding connections, or an invalid LoRA configuration, may cause a failure. The provided source does not specify the resulting exception text.
- Solution: Check the LoRA configuration and referenced stacks. If you are not intentionally using legacy sidecars, disconnect both
lora_registryandlora_bindingsso the node uses the regional context’s LoRA scopes.
Regional guidance is absent or weaker than expected
- Explanation: This is an output-quality symptom rather than a specified error message. A low
attention_strength, a narrow sampling interval, or regional instructions that do not express the intended areas and prompts can make the effect difficult to see. - Solution: Confirm the regional setup first, then compare a run using the defaults—
1.0strength,0.0start, and0.5end—with runs that change only one setting at a time.
Sampling interval is reversed or has no duration
- Explanation: The interface permits both percentages from
0.0to1.0; the shown node code does not validate their order. An end value at or before the start does not describe a useful forward interval. - Solution: Set
start_percentlower thanend_percent, such as the defaults0.0and0.5.
