SD1.5x Dynamic Super Resolution · सार्वजनिक Open Skill
SD1.5x Dynamic Super Resolution
SD1.5x Dynamic Super Resolution Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Product
3 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/vhjy4wxj.jpg
# SD1.5x Dynamic Super Resolution Turns one uploaded picture into one high-resolution image. The source subject, structure and composition are kept, while the whole frame is re-rendered with the SD1.5x dynamic super-resolution look: SD1.5 backbone quality, multi-LoRA detail enhancement, image magnification, lossless sharpening, and stylized rendering of complex scenes. This is not a generic "generate an image" skill — every result must be a recognizable high-fidelity upgrade of the input image. There is no structured project state. Continuity lives in the conversation, in the assets already on the canvas, and in the todo list. ## When to use - The user uploads an image and asks to enlarge, upscale, sharpen, enhance, clean up, or "make it HD". - The user wants their picture re-rendered in the SD1.5x dynamic super-resolution style (stylized detail enhancement, richer light and material). - The user wants a complex scene from their own image regenerated at high resolution with more micro-detail. - The user asks to fix, redo, or push further a result this skill produced earlier in the conversation. ## Inputs and outputs - Input: a reference image (required for the main path) plus optional text that names subject, mood, palette, text elements, or desired framing. - Output: one image asset. Name it with a short human title and use `entity_type='final_images'`. ## Steps 1. **Confirm the inputs.** Find the reference image in the conversation or on the canvas. If the user gave only text, ask once for the image to enhance. If they insist on text-only, follow the text-only branch in `references/enhancement-pipeline.md` and use `model='t2i'`. 2. **Read the visual rules.** Call `read_skill_file` with `references/style-and-prompt.md` before writing any prompt. It holds the Style, Subject/content, Layout, Must keep, Must avoid blocks and the reusable English prompt templates. 3. **Read the operating steps.** Call `read_skill_file` with `references/enhancement-pipeline.md`. It holds the intake checklist, the optional image inspection call, the orientation/resolution decision table, prompt assembly order, pre-call validation, post-call QC and the retry rules. 4. **Inspect the source when detail matters.** If the image is busy (characters, small text, logos, vehicles, machinery) or the user stresses identity fidelity, call `recognize_image` with the inspection prompt from `references/enhancement-pipeline.md` to list the must-keep elements. Do not skip this when the user says "keep it exactly the same". 5. **Write the prompt.** Assemble an English prompt with the template and fragments in `references/style-and-prompt.md`, filled with the concrete subject, scene, materials, light, palette and text elements from this specific image and this specific request. No vague filler, no copy-paste boilerplate as the final prompt. 6. **Generate.** Call `generate_image` with: - `prompt`: the assembled English prompt, - `input_urls=[reference image]` (asset id, `s3://` uri, preset key, or the URL returned earlier in this conversation), - `model='i2i.hd.lite'`, - `orientation` and `resolution` chosen with the decision table in `references/enhancement-pipeline.md` (match the source aspect ratio; default `1080P`, `2K` when the user asks for maximum detail), - `name`: short title such as `Cat Astronauts HD`, - `entity_type='final_images'`. 7. **Check the result** against the QC list in `references/enhancement-pipeline.md`. If identity, layout, or text broke, run at most one corrective pass with a tightened prompt; never loop more than that. 8. **Reply** in natural language: what was enhanced, what was preserved, and the practical next options. Keep URLs out of the reply body. If the work is finished, call `set_guidance` in the same turn with 2–3 concrete follow-ups (e.g. sharper micro-detail, 2K pass, alternate palette). ## Notes - The `.hd` image channels do not support NSFW or adult content. Decline such requests politely and do not