The max_batch setting was being ignored with anima previews, had a LLM debug and it fix it, it spat out this:

Bug: max_batch ignored for video model latent formats when video_max_frames is -1

Description

The max_batch setting in betterTaesdPreviews (or previews) is ignored when using a video model latent format (e.g., wan21), even when doing batch image generation (not video). The preview always shows all batch items regardless of max_batch.

Root Cause

In py/better_previews/previewer.py:calculate_indexes(), when is_video=True (5D latent tensor), the code uses SETTINGS.btp_video_max_frames as the limit instead of self.max_batch_preview:

def calculate_indexes(self, batch_size: int, *, is_video=False) -> range:
    max_batch = (
        SETTINGS.btp_video_max_frames if is_video else self.max_batch_preview
    )
    if max_batch < 0:
        return range(batch_size)  # ← video_max_frames=-1 returns ALL items
    ...

When video_max_frames is -1 (default = unlimited), the condition max_batch < 0 is True, so range(batch_size) is returned — showing all batch items regardless of max_batch.

Steps to Reproduce

1. Use any video model latent format (e.g., WAN21, HunyuanVideo, etc.) — the latent is 5D [batch, channels, frames, H, W]
2. Set max_batch: 4 (or 2) in blehconfig.yaml
3. Leave video_max_frames: -1 (the default, unlimited)
4. Set batch_size: 5 in the KSampler
5. Observe: the live preview shows 5 items in the grid, ignoring max_batch

Expected Behavior

When video_max_frames is -1 (unlimited), the batch limit should fall back to max_batch_preview. That is, calculate_indexes should use max_batch_preview to limit batch items when video_max_frames doesn't provide a limit.

Proposed Fix

In calculate_indexes, when max_batch < 0 and is_video=True, fall back to self.max_batch_preview instead of returning all items:

def calculate_indexes(self, batch_size: int, *, is_video=False) -> range:
    max_batch = (
        SETTINGS.btp_video_max_frames if is_video else self.max_batch_preview
    )
    if max_batch < 0:
        if is_video:
            max_batch = self.max_batch_preview  # fall back to batch limit
        else:
            return range(batch_size)
    if not self.maxed_batch_step_mode:
        return range(min(max_batch, batch_size))
    return range(
        0,
        batch_size,
        math.ceil(batch_size / max_batch),
    )[:max_batch]

Workaround

Set video_max_frames to a specific value (e.g., video_max_frames: 4) instead of -1. This forces a limit on the frame/batch dimension for video models. However, this conflates frame limiting with batch limiting, which may not be desired.