Kling v3 Omni (BudgetPixel)

Kling v3 Omni is the Kling image family's answer to "what if we could hold a whole character sheet at once?" It's a unified text-to-image and image-editing node with multi-reference composition from up to nine reference images and output that goes up to 4K. In the BudgetPixel image lineup that puts it in the same heavy-lifting company as GPT-Image-2 and FLUX 2 Dev - the nodes you reach for when a single reference isn't enough to pin down who, what, and where.

The pricing is where Omni gets interesting: 35 credits per image at 1K/2K, and 60 at 4K. The 4K step-up is only 25 credits over the 2K price - proportionally tiny compared to the video shelf - which makes the 4K tier genuinely tempting. "Omni" is the pitch: one node that does generation and editing with serious reference capacity, without the billing matrix complexity of the OpenAI or Grok 2.0 nodes.

How it works

Standard BudgetPixel image flow: POST your prompt and reference batch into the Kling image pipeline, poll /images/{id}, download the IMAGE batch. The reference handling is the star - batches you wire into the reference_images input map onto the model's reference slots automatically, so a batch of nine images becomes nine composition anchors with zero extra wiring. That's the same pattern as FLUX 2 Dev but with more than double the reference capacity.

The inputs that matter

• prompt - required, multiline. With references attached, this is the composition instruction: "image 1 is the subject, image 2 the outfit, image 3 the location…"
• reference_images - up to 9. The node's whole reason to exist.
• size - 1K / 2K / 4K. The 4K tier is only 60 credits - the cheapest high-res path in the pack's image lineup.
• aspect_ratio - framing.
• num_images - 1–4, default 1 (each billed at the tier's rate).
• seed - default -1 (random).

Output is images (IMAGE batch).

Install

Manager (search "BudgetPixel") or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Restart, set BUDGETPIXEL_API_KEY or budgetpixel.json ({"api_key": "bpx_live_..."}) in the pack folder. No model files; requests is the only dependency.

Where people get burned

Reference hygiene is everything at nine slots. Feeding it nine images that conflict - different people, different lighting, contradictory styles - produces a muddle, and the 35-60 credits you spent aren't coming back. Curate the batch like it's a casting call. Second, remember this is Kling's image model despite the Kling video brand - if you came here expecting video, the video Kling nodes are in the BudgetPixel/Video category. Third, the 4K tier's cheap price can lead to overshooting: a 4K Omni render is slower server-side and heavier downstream, so use 4K when you'll actually use the resolution, not because it's only 25 credits more. And when you re-test, flip force_rerun so the cache doesn't hand you yesterday's image.