Atlas Image API X

Atlas Image API X is WorkflowX's wrapper for the Atlas reseller service, and on the surface it looks nearly identical to its sibling Kie Image API X - same polling pattern, same quality dials, same "pick a closed model from a catalog, prompt it, get an IMAGE back." The differences are in the catalog and the size control, and those are worth understanding before you pick between the two.

The model list is its own mix of closed and hosted families: wan-2-7 and wan-2-7-pro, gpt-image-1-5, gpt2, the seedream-5 line (lite/pro plus seedream-4-5), nano-banana-2 and nano-banana-pro, qwen2, kontext, and qwen-image-2-pro. If your workflow wants Nano Banana 2 on Monday and Seedream 5 Pro on Tuesday, this node is a single dropdown's worth of switching - no new SDK, no new node.

The thing that actually sets Atlas apart for output control is image_size, which has 54 choices spanning named tiers (1K/2K) and explicit pixel dimensions (1024×1024, 1024×1536, 2048×2048, 3840×2160, 2160×3840, and more) plus an auto mode. That granularity matters when you're targeting a specific canvas, video frame, or print size instead of just "bigger." For anything off the list, custom_size_enabled with custom_width/custom_height (64–8192) covers it.

The mechanics

Like the Kie node, this one is asynchronous: submit the generation, then poll until the result is ready. timeout_seconds (default 600), poll_interval_seconds (default 3), and retrieval_mode (generate vs force_retrieve for pending tasks) drive that loop. The shared input surface is the WorkflowX API contract: api_key (optional, falling back to ATLAS_API_KEY or ATLASCLOUD_API_KEY), prompt, aspect_ratio, quality/guidance/steps dials, safety checkers and watermark, seed_enabled+seed, thinking_mode, web/image search toggles, and show_payload for debugging. Optional image_1 (IMAGE) provides a reference, with the frontend adding more sockets as you connect them.

Output: a single image (IMAGE) on a normal socket, so the result feeds your existing preview and save nodes.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/haroonaslam/WorkflowX-Configurator

or search WorkflowX Configurator in ComfyUI Manager, restart, refresh. No pip deps, no model downloads.

The honest tradeoffs (mostly shared with Kie)

• It's a middleman. Reseller pricing, region access, and billing convenience come with inputs leaving your machine and per-call cost. The vendor model enforces its own moderation regardless of what the reseller advertises.
• Size granularity is a feature and a trap. 4K and 3840×2160 calls cost more and take longer - draft in 1K, then hit the big sizes only for finals. A 54-option dropdown makes it easy to accidentally order a 4K render for a test.
• Polling blocks its branch. Each call runs until the provider returns or timeout_seconds expires. In a shared workflow, that's expected behavior, not a hang.
• Environment keys beat widget keys. Set ATLAS_API_KEY and share workflows without serializing credentials into the JSON.

Between Atlas and Kie, pick by catalog and size needs - Atlas wins when pixel-exact output sizes matter, Kie wins if the model you need is on its roster. Both are the same idea: closed models inside your graph, with the API tax fully visible.