WebP files don't work for me

[mcjean-netizen]

WebP files don't work for me, and all my work is in WebP format. The few PNG ones I have work fine with no problems. Everything was made in ComfyUI and saved with metadata; I can even reload them into the canvas and the full workflow shows up perfectly. I've tried them on various websites too.

[erroralex]

Hi, thanks for the report.

I just released version 1.0.8, which explicitly fixes WebP support. Previously, WebP files were being filtered out by the application, but they should now load and display correctly along with their metadata.

Please give 1.0.8 a try and let me know if it solves the issue for you!

[mcjean-netizen]

The issue has been addressed, but not resolved.

The topic isn't closed; you should ask before closing a topic.

Real prompt:

(German Black Forest valley:1.3), (dense spruce, fern floor, light fog), (deep greens), Cozy:, You are a master photographer who specializes in chiaroscuro, using a single intense hard white light source to generate dramatic light and shadow.

Your tastes encompass brilliant glowing whites, with sharp black shadows, and extreme contrast with white burned areas.

---

Take the following erotic photograph with adult content and sexual provocation:

You are a seasoned cinematographer who works with intense side lighting and gobo‑crafted patterns to sculpt deep, sharply defined shadows.

Your aesthetic embraces a low‑key, high‑contrast chiaroscuro look: dramatic contrasts rendered in a muted‑yet‑vibrant Ektachrome‑inspired palette with a subtle warm tint, and a slightly desaturated finish that evokes timeless, cinematic noir.

---

Take the following indoors erotic photograph with adult content and sexual provocation:

Negative prompt:

色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走

[erroralex]

The issue should now be resolved because the ComfyUIStrategy has been updated to correctly trace text connections through complex custom routing nodes, which were previously breaking the link between your prompt text and the Sampler.

Why it failed before

In your workflow, the "Real Prompt" (starting with (German Black Forest valley:1.3)...) flows through a very deep and complex chain of logic nodes before reaching the Sampler. Specifically, it passes through:

1. JPS_DynamicPromptConcatenate (Node 100 & 108)
2. Text Concatenate (JPS) (Node 53)
3. StringReplace (Nodes 49, 50, 51, 52)
4. Any Switch (rgthree) (Node 54)
5. Reroute (Node 43)

Previous versions of the parser likely stopped tracing when they encountered these specific custom nodes (like "Switch" or "Concatenate"), causing the parser to lose track of where the text was going. As a result, it couldn't confirm the text was used for generation, often leading it to either ignore the text or dump it as "Unknown/Junk" alongside unrelated nodes.

Why it works now

The updated traceNodeOutputs method in ComfyUIStrategy.java now explicitly handles these routing types. It can "see through" them to find the destination Sampler:

// From the updated ComfyUIStrategy.java
if (destType.contains("reroute") || destType.contains("pipe") || destType.contains("bus")
        || destType.contains("switch") || destType.contains("concatenate") // <--- Handles JPS nodes
        || destType.contains("replace") || destType.contains("processor")  // <--- Handles StringReplace
        || destType.contains("string") || destType.contains("text")) {
    // It now recurses deeper to find the connection
    String res = traceNodeOutputs(dest, linkDestMap, depth + 1);
    if ("Negative".equals(res) || "Positive".equals(res) || "Prompt".equals(res)) return res;
}

The Result

Because the parser can now successfully trace the path from your Primitive Nodes (ID 71, 74, etc.) all the way to the ClownsharKSampler_Beta (Node 46), it correctly identifies that text as the Positive Prompt.

1. Positive Prompt: Correctly identified as the text driving Node 46 (via the Switch/Concatenate chain).

2. Negative Prompt: Correctly identified from CLIPTextEncode (Node 26) which connects directly to the Sampler's negative input.

Let me know if the updated V.1.0.9 works for your specific workflow.

[mcjean-netizen]

This workflow contains nodes coded by me—you won’t be able to reproduce it exactly as-is, but that shouldn’t stop you from focusing on what you actually need.

No matter how complex a workflow may become—even if, as in this case, it involves more than 45 windows containing text—your program only needs to display one: the one connected to (or feeding into) node 40, the “CLIP Text Encode (Prompt)” node.

Additionally, the data your program shows for the KSampler is incorrect—it pulled information from a KSampler that was in Bypass mode. It should always fetch data from the active sampler(s) (there can be multiple). In this case, there’s only one active, and it’s node 46.