invoke-ai/InvokeAI · error · ValueError

Negative text conditioning is required when cfg_scale is not

Error message

Negative text conditioning is required when cfg_scale is not 1.0.

What it means

In Flux denoise_sampling_region (denoise.py), when the per-step CFG scale is not 1.0 the sampler must also run a negative (unconditional) prediction, which requires negative text conditioning supplied via a neg_regional_prompting_extension. When a step's cfg_scale != 1.0 and that extension is None, the sampler cannot form the negative pass and raises. CFG == 1.0 means no negative branch is needed, hence the conditional check.

Source

Thrown at invokeai/backend/flux/denoise.py:177

                    timesteps=t_vec,
                    guidance=guidance_vec,
                    timestep_index=user_step,
                    total_num_timesteps=total_steps,
                    controlnet_double_block_residuals=merged_controlnet_residuals.double_block_residuals,
                    controlnet_single_block_residuals=merged_controlnet_residuals.single_block_residuals,
                    ip_adapter_extensions=pos_ip_adapter_extensions,
                    regional_prompting_extension=pos_regional_prompting_extension,
                )

                if img_cond_seq is not None:
                    pred = pred[:, :original_seq_len]

                # Get CFG scale for current user step
                step_cfg_scale = cfg_scale[min(user_step, len(cfg_scale) - 1)]

                if not math.isclose(step_cfg_scale, 1.0):
                    if neg_regional_prompting_extension is None:
                        raise ValueError("Negative text conditioning is required when cfg_scale is not 1.0.")

                    neg_img_input = img
                    neg_img_input_ids = img_ids

                    if img_cond is not None:
                        neg_img_input = torch.cat((neg_img_input, img_cond), dim=-1)

                    if img_cond_seq is not None:
                        neg_img_input = torch.cat((neg_img_input, img_cond_seq), dim=1)
                        neg_img_input_ids = torch.cat((neg_img_input_ids, img_cond_seq_ids), dim=1)

                    neg_pred = model(
                        img=neg_img_input,
                        img_ids=neg_img_input_ids,
                        txt=neg_regional_prompting_extension.regional_text_conditioning.t5_embeddings,
                        txt_ids=neg_regional_prompting_extension.regional_text_conditioning.t5_txt_ids,
                        y=neg_regional_prompting_extension.regional_text_conditioning.clip_embeddings,
                        timesteps=t_vec,

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Provide a negative regional prompting extension (with negative text embeddings) whenever cfg_scale != 1.0.
2. Keep cfg_scale at 1.0 (guidance-distilled Flux models are trained to run without CFG) if no negative conditioning is available.
3. Build the negative prompt embeddings in your pipeline before denoising and pass them through.
4. If using a per-step cfg_scale list, ensure every entry that exceeds 1.0 is accompanied by negative conditioning, or clamp them to 1.0.

Example fix

// before
result = denoise(..., cfg_scale=4.0, neg_regional_prompting_extension=None)
// after
neg_ext = build_neg_regional_prompting_extension(negative_prompt="blurry, low quality", ...)
result = denoise(..., cfg_scale=4.0, neg_regional_prompting_extension=neg_ext)

if any(s != 1.0 for s in (cfg_scale if isinstance(cfg_scale, list) else [cfg_scale])):
    assert neg_regional_prompting_extension is not None, \
        "cfg_scale != 1.0 requires negative text conditioning"

def cfg_needs_negative(cfg_scale, neg_ext) -> bool:
    values = cfg_scale if isinstance(cfg_scale, list) else [cfg_scale]
    return any(s != 1.0 for s in values) and neg_ext is None

try:
    result = denoise(..., cfg_scale=cfg_scale, neg_regional_prompting_extension=neg_ext)
except ValueError as e:
    if "Negative text conditioning is required" in str(e):
        raise RuntimeError("Enable CFG only with a negative prompt / conditioning extension configured") from e
    raise

When it happens

Trigger: Calling denoise() with a cfg_scale (scalar or per-step schedule) whose value at some step is not 1.0 while neg_regional_prompting_extension is None — e.g. passing cfg_scale=3.5 without building negative conditioning.

Common situations: Users enabling CFG in a Flux pipeline that was set up without a negative prompt/region extension; per-step CFG schedules where a later step exceeds 1.0; region-based prompting setups lacking the negative extension object.

Related errors

• Negative text conditioning is required when cfg_scale is not
• denoising_start ({self.denoising_start}) must be less than d
• denoising_start should be 0 when initial latents are not pro
• cfg_scale must be greater than 1
• Negative conditioning is required when guidance_scale > 1.0

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/13439fcacfd449ac. Report an issue: GitHub.