Why copying nodes is the wrong migration
ComfyUI is powerful because it lets advanced creators wire almost anything together. That same flexibility can become the problem. You start with a simple goal like "make my original character walk through a rainy alley," then the workflow turns into model loaders, reference branches, masks, pose preprocessors, video adapters, upscalers, samplers, reroute nodes, preview nodes, and a pile of saved versions named almost the same thing. The pain is not that node tools are bad. The pain is that the minimum path to a usable anime shot becomes hard to see.
ArcLoop shouldn't be treated as a visual clone of a node graph. The better migration is conceptual. Keep the parts of the workflow that actually protect the result: the character asset, the reference type, the shot goal, the camera, the action, the draft pass, and the review checklist. Drop the parts that only exist because a local graph needs manual plumbing.
This guide is for ComfyUI users who already understand references, prompts, and iteration, but want a shorter route for anime shorts, OC scenes, storyboard tests, and repeatable Explore templates. It focuses on the smallest migration that works: one character asset, one reference decision, one shot brief, one low-cost draft, one refinement pass. The ComfyUI Migration Mini Flow is the matching Explore workflow.
Decisions that replace the node graph
The first principle is to migrate intent, not nodes. If a node only existed to pass image size or condition one branch, it may not need a direct ArcLoop equivalent. If a node protected identity, pose, composition, timing, or final quality, preserve that purpose in the new workflow.
The second principle is to reduce the number of moving parts per shot. ComfyUI makes it tempting to solve style, character, pose, camera, background, upscale, and edit in one graph. In ArcLoop, treat the shot as a production brief. Decide what must stay fixed and what is allowed to change.
The third principle is reference choice before prompt length. Many hard workflows exist because the user is forcing one input to do too many jobs. A character sheet should lock identity. An outfit reference should protect clothing. A pose reference should control body language. A scene reference should guide layout and depth.
The fourth principle is draft before polish. A cheap rough pass is not a downgrade. It is how you avoid spending final-quality budget on a shot with the wrong action, wrong framing, or broken continuity.
The fifth principle is readable review. A node graph can hide decisions in cables. A migrated workflow should leave reviewable fields: character, world, shot goal, camera, action, references, negative constraints, and output notes.
How to migrate a shot step by step
Start by naming the output, not the tool stack. Write one sentence: "Create an 8-second anime shot of this OC entering a neon alley and noticing a signal on her wrist device." This keeps the migration tied to a result instead of a graph shape.
Next, list the nodes that actually affect creative control. Ignore utility nodes at first. Keep anything that defines identity, reference image, pose, depth, camera movement, model style, prompt text, timing, or quality pass.
Translate those nodes into ArcLoop decisions. A face reference becomes the character identity field. An outfit branch becomes clothing lock notes. A pose or depth extractor becomes action and camera guidance. A sampler experiment becomes a draft-versus-final decision, not a permanent ritual.
Build the minimum flow: character sheet, reference selection, shot prompt, draft generation, review, final generation. Do not start with a giant reusable system. Run one shot through the path and only keep the steps that changed the result.
Write the shot prompt as a small production brief. Include subject, setting, action, camera, style, and constraints. If the old graph used multiple references, specify why each one exists. The goal is not fewer words at all costs. The goal is fewer hidden dependencies.
Review with a short checklist. Did identity hold? Did the action happen? Did the camera stay readable? Did the reference solve the right problem? Did the draft expose a flaw before final spend? Use Low-Cost Draft to Final for this gate.
What's worth keeping from your ComfyUI habits
Keep the discipline of separating inputs. Advanced ComfyUI users already know that one reference can push identity while another pushes pose. That habit is valuable. Bring it into ArcLoop by naming reference roles instead of burying them in node paths.
Keep the habit of inspecting intermediate results. If a rough image or draft clip already shows that the character is wrong, do not polish it. The cheapest migration keeps checkpoints, but removes needless wiring around them.
Keep technical specificity when it helps the image. Camera direction, pose clarity, background depth, negative constraints, and output duration still matter. What changes is where you write them. They move into reviewable prompt sections and template fields.
Keep version awareness, but simplify it. You do not need ten graph copies for a one-shot test. Save one working template, one character sheet, and one final shot note. Add complexity only when a real repeated project proves it needs it.
Example 1: Character Lock Migration
ArcLoop workflow brief: migrate a node-heavy character consistency test into a simple character asset first pass.
Create a clean 2D anime character sheet for an original courier named Lina. She has short silver hair with two black hairpins on the left side, amber eyes, a cropped yellow jacket, black cycling shorts, white roller boots, and a blue delivery sling bag with a star patch. Generate a front view, side view, back view, and one small neutral face close-up on a plain light background. Keep the jacket shape, hairpins, roller boots, and sling bag visible in every view. Use crisp cel-shaded anime line art, readable proportions, simple clothing folds, no background scene, no action pose, no text, no logo, no extra fingers.
This prompt replaces a messy identity workaround with the asset the workflow needed in the first place: a stable sheet.
Example 2: Shot Prompt Migration
ArcLoop shot prompt: use the Lina character asset as identity reference (`@Lina`). Use a rainy cyberpunk alley background reference only for depth and lighting, not for character design.
Create an 8-second 2D anime video shot. Lina, the original courier with short silver hair, black hairpins, yellow cropped jacket, white roller boots, and blue sling bag, skates slowly into a narrow neon alley after rain. Main action: she stops beside a vending machine, looks down as the star patch on her bag begins to glow, then turns her face toward camera with sudden concern. Camera: medium side-tracking shot that becomes a locked three-quarter view at the stop. Secondary motion: puddle reflections ripple under the roller boots, and neon signs flicker across her jacket. Preserve her face, hairpins, jacket, boots, and bag. No redesign, no extra limbs, no text, no watermark, no photorealism.
The prompt names which reference controls which problem. That is the part worth migrating from a graph.
Example 3: Draft-to-Final Migration
ArcLoop two-pass workflow:
Draft pass: generate a low-cost 5-second rough version of Lina entering the alley. Use simple lighting, one side-tracking camera, and one action only: she skates in, stops, and notices the glowing star patch. Evaluate only identity, framing, and action readability.
Final pass: if the draft works, regenerate as an 8-second polished 2D anime scene with wet asphalt reflections, magenta and teal neon depth layers, soft rain, subtle jacket motion, and a stronger concerned expression at the end. Keep the same character sheet reference and the same camera path. Do not add new actions during the final pass. Preserve the yellow jacket, black hairpins, white roller boots, and blue sling bag. No text, no logo, no extra limbs, no sudden outfit change.
This is often cheaper than chasing a perfect first render. The draft proves the shot before the expensive polish.
Where migrations break most often
The first mistake is trying to rebuild the whole graph as a product template. A graph may contain dozens of nodes, but the actual creative contract is usually much smaller: who, where, what action, which reference, what output.
Another mistake is dropping the reference logic. If the old workflow used separate identity, pose, and scene control, do not compress them into one vague prompt. Simplification should remove wiring, not remove control.
Creators also migrate too much quality work too early. Upscale, high resolution, long duration, and heavy effects should wait until the draft proves the shot.
A fourth mistake is hiding decisions in file names. A reusable ArcLoop workflow should be readable by another person without opening a node graph. Name the character, reference role, shot goal, camera, action, and constraints.
Finally, do not promise that every advanced local graph becomes one click. Some projects need custom tooling. The lazy migration is to move the repeatable 80% first, then keep specialist work for shots that actually need it.
FAQ
Do I need to recreate every ComfyUI node in ArcLoop?
No. Recreate the production intent: identity, reference role, shot action, camera, draft pass, and final quality pass. Utility wiring does not need a one-to-one replacement.
What should I migrate first?
Start with one repeatable anime shot. Build a character sheet, choose references by role, draft a short clip, and refine only after the draft is readable. Do not begin with a giant library.
What if my ComfyUI workflow has an 85% success workaround?
Keep the lesson, not the workaround. If the workaround succeeds because it creates a neutral input, masks the generation area, and protects identity, turn those ideas into a character sheet plus reference-control workflow.
