invoke-ai/InvokeAI · error · ValueError

{self.__class__} has the config param `addition_embed_type`

Error message

{self.__class__} has the config param `addition_embed_type` set to 'text_image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`

What it means

The UNet config sets `addition_embed_type='text_image'` (Kandinsky 2.1 style extra conditioning), so `forward()` requires `added_cond_kwargs={'image_embeds': ...}` to build the augmentation embedding from text+image embeddings. When `image_embeds` is absent from `added_cond_kwargs`, the model raises instead of silently producing wrong conditioning.

Source

Thrown at invokeai/backend/hidiffusion/hidiffusion.py:1078

                    class_labels = self.time_proj(class_labels)

                    # `Timesteps` does not contain any weights and will always return f32 tensors
                    # there might be better ways to encapsulate this.
                    class_labels = class_labels.to(dtype=sample.dtype)

                class_emb = self.class_embedding(class_labels).to(dtype=sample.dtype)

                if self.config.class_embeddings_concat:
                    emb = torch.cat([emb, class_emb], dim=-1)
                else:
                    emb = emb + class_emb

            if self.config.addition_embed_type == "text":
                aug_emb = self.add_embedding(encoder_hidden_states)
            elif self.config.addition_embed_type == "text_image":
                # Kandinsky 2.1 - style
                if "image_embeds" not in added_cond_kwargs:
                    raise ValueError(
                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
                    )

                image_embs = added_cond_kwargs.get("image_embeds")
                text_embs = added_cond_kwargs.get("text_embeds", encoder_hidden_states)
                aug_emb = self.add_embedding(text_embs, image_embs)
            elif self.config.addition_embed_type == "text_time":
                # SDXL - style
                if "text_embeds" not in added_cond_kwargs:
                    raise ValueError(
                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
                    )
                text_embeds = added_cond_kwargs.get("text_embeds")
                if "time_ids" not in added_cond_kwargs:
                    raise ValueError(
                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
                    )
                time_ids = added_cond_kwargs.get("time_ids")

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Pass `added_cond_kwargs={'image_embeds': image_embeds}` to the UNet forward call
2. Pass `text_embeds` too if the text embeddings differ from `encoder_hidden_states` (optional; it defaults to encoder_hidden_states)
3. If the checkpoint is not Kandinsky-style, load a UNet with `addition_embed_type=None`

Example fix

// before
noise_pred = unet(sample, t, encoder_hidden_states=text_emb)
// after
noise_pred = unet(sample, t, encoder_hidden_states=text_emb,
                  added_cond_kwargs={'image_embeds': image_embeds})

if getattr(unet.config, 'addition_embed_type', None) == 'text_image' and not (added_cond_kwargs and 'image_embeds' in added_cond_kwargs):
    raise ValueError("text_image UNet requires added_cond_kwargs={'image_embeds': ...}")

def has_image_embeds(added_cond_kwargs) -> bool:
    return isinstance(added_cond_kwargs, dict) and 'image_embeds' in added_cond_kwargs

try:
    out = unet(sample, t, emb, added_cond_kwargs=ackw)
except ValueError as e:
    if 'image_embeds' in str(e):
        ackw = {'image_embeds': image_embeds}; out = unet(sample, t, emb, added_cond_kwargs=ackw)
    else: raise

When it happens

Trigger: Calling forward on a UNet whose `config.addition_embed_type == 'text_image'` (Kandinsky 2.1) without passing `added_cond_kwargs` or passing a dict that lacks the `image_embeds` key.

Common situations: Loading a Kandinsky 2.1 checkpoint and calling the UNet directly with only sample/timestep/text embeddings; adapting an SD pipeline loop to Kandinsky without adding the extra kwargs; copying UNet forward calls between models with different `addition_embed_type` configs.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

• {self.__class__} has the config param `addition_embed_type`
• {self.__class__} has the config param `encoder_hid_dim_type`
• class_labels should be provided when num_class_embeds > 0
• {self.__class__} has the config param `addition_embed_type`
• {self.__class__} has the config param `addition_embed_type`

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/ebb5739e172f3459. Report an issue: GitHub.