Three usage notes across two pages.

Qwen Image 2.1 (tutorials/image/qwen/qwen-image-2-1.mdx):

• Prompt enhancer max_length budget. All four templates ship it at 4096, and the enhancer model cards sample at 16256 (T2I) and 24000 (I2I) with thinking on. When the reasoning block exhausts 4096 the rewrite comes back empty, Preview Any shows nothing, and the image model samples that empty prompt. Also corrects "both templates" to all four and adds max_length to the control list.
• New section "Faster sampling with the Turbo LoRA": the official Turbo LoRA shipped in the model repository, run at 8 steps and cfg 1 with LoraLoaderModelOnly on the base checkpoint, so no second diffusion model download. The Lightning section now sends readers to the Qwen Image 2.1 subgraph node for steps, which is where both base templates promote that widget, instead of KSampler.

MOSS-Image (tutorials/image/moss-image/moss-image.mdx):

• Layout workflow: how boxes are numbered and labelled, the character_boxes field (pre-filled 2 in SFT, 1 in Base), what the JSON prompt contains, and why enable_prompt_enhancer is off there.
• New "Going above 2 MP": render at 1 MP to 2 MP, resize in pixel space with Upscale Image, encode, then sample again at partial denoise. Repeat instead of jumping to the final size; an 8K canvas needs VAE Decode (Tiled).
• New "Reference images and the prompt upsampler": ref_image (for llm) feeds the upsampler only, because the model has no image input.
• Renamed the switch to its actual label enable_prompt_enhancer, corrected the resolution tip to the recommended output range, and dropped the "SLG is on by default" claim, which does not hold for every template.

Basis: node behaviour from comfy_extras/nodes_moss_image.py, nodes_textgen.py, nodes_bounding_boxes.py; the Qwen Image 2.1 model card for the Turbo LoRA setting (8 steps, cfg 1); promoted labels and widget values from the image_qwen_image_2_1_{t2i,image_edit,turbo_t2i} and image_moss_image_* template JSONs; sampling and resolution guidance from the official MOSS-Image USERGUIDE.

zh/ja/ko translations are in the same commits.