Concurrent Submit | Zhenzhen_qwen_image_edit

The editing half of Zhenzhen's Qwen image pair. ComflyConcurrent_Comfly_qwen_image_edit_Submit wraps Alibaba's qwen-image-edit model, which takes an input image plus a text instruction and returns an edited version - change the background, swap an object, re-style a scene. The concurrent wrapper is where it earns its keep: image edits are exactly the job you want to batch, and this node lets you queue a stack of them without the graph freezing on each one.

Mechanically it's the same pattern as every Submit node in this pack. The node runs in the shared image pool (30 workers), returns a COMFLY_IMAGE_FUTURE task instead of blocking, and you collect up to 30 tasks with Concurrent Collect Images (ComflyConcurrent_Image_Await), which waits for all of them and hands back image_1…image_30 in slot order. Submit first, collect later, stay responsive.

The inputs

• prompt - the edit instruction ("make the sky sunset orange," "replace the bottle with a vase").
• image - the source image. This is required, unlike the t2i twin - it's an edit node, there's nothing to edit without one.
• size / Custom_size - preset resolutions or your own 1280x720-style value. Custom resolution support is a genuine Qwen advantage.
• model - locked to qwen-image-edit.
• num_images - 1 to 4 variations of the edit. Again, billing scales with image count.
• acceleration - none / regular / high. The speed-vs-quality dial; high is faster and cheaper, none is the careful pass. This is a nice knob most API packs don't expose.
• num_inference_steps (2–50), guidance_scale (0–20), negative_prompt, enable_safety_checker, output_format (png/jpeg), seed, apikey, skip_error - the rest.

Output is the single task (COMFLY_IMAGE_FUTURE). Don't skip the Await step.

Install

Same drill as the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Or ComfyUI Manager → search Comfyui-zhenzhen → restart. Token from ai.t8star.org (overseas) or api.seedance.nz (domestic discount) into apikey.

Where people trip up

• The default group is the slow lane for Qwen edits. The README's advice for Google models - switch to a quality group when the default drags - applies here too; the reseller routes through token groups you can change on the site.
• acceleration: high changes the look, not just the speed. If your edits suddenly look rougher, this is why. Only reach for high when you're testing or you genuinely don't care about the extra quality.
• Custom size flakiness. Qwen doesn't always honor arbitrary resolutions; if output comes back at a preset, that's the endpoint, not your typing. And as always with this pack: a timeout doesn't mean failure - check the async-task page before resubmitting and paying twice.