A Recolor and Colorway-Variant Workflow

A colorway variant is not simply a hue-shifted copy of a source image. It must preserve material texture, highlight placement, shadow depth, and any surface imperfections — changing only the dominant hue and saturation of the target object while leaving the background, lighting, and form model untouched. Standard photo-editing hue sliders shift every pixel in the image uniformly and cannot distinguish the product surface from its environment. AI recoloring nodes, by contrast, use semantic understanding of the scene to isolate the target object and resynthesize its surface in the new color.

This semantic distinction is why the workflow matters. When you build the recolor pipeline in the Floniks editor, you are encoding a series of decisions: which object receives the new color, what descriptors define the target hue, how much denoising latitude to give the model, and how to validate that the output matches your brand palette. Running these decisions as an explicit node graph makes them auditable and repeatable — critical when you need to produce 8 SKU colorways from a single hero product shot before a launch deadline.

The variant workflow also enables parallelism. Instead of running one recolor job, waiting, inspecting, and running the next, you branch the workflow into N parallel tracks — one per colorway — and collect all results in a single output node. This means the full palette render takes approximately the same wall-clock time as a single recolor.

Floniks offers two distinct approaches to recoloring. The first is a dedicated Color Transfer node that accepts a source image and a target color swatch or reference image and applies the dominant palette of the reference to the source while preserving luminance structure. This works well when you have an approved brand swatch as your target. Connect the product image to the Image input, and connect a single-color swatch (a 100x100 pixel block of the exact hex color) to the Reference input. Set influence strength to 0.7–0.8 to allow the model to adapt tone and saturation to the object geometry without over-imposing the flat swatch.

The second approach uses an image-to-image node with a color-directive prompt. This is more flexible when the target color needs contextual nuance — for example, a navy blue that reads differently on knit fabric versus smooth leather. Write the prompt as: "navy blue [object], same surface texture, same lighting direction, same background, photorealistic, high detail." Set denoising strength to 0.5–0.65. Too high and the model regenerates geometry; too low and the color shift is minimal. The sweet spot for most product materials is 0.55.

For objects with complex material interactions — metallics, translucent fabrics, iridescent finishes — use the color-directive prompt approach because the Color Transfer node can misattribute spectral highlights as hue rather than luminance, producing an output where specular reflections are tinted rather than white. In the prompt, explicitly state "metallic specular highlights remain neutral white" or "iridescent sheen consistent with new base color" to guide the model correctly.

The most common failure mode in AI recoloring is texture drift: the model changes the color but also smooths out surface texture, removes embossed logos, or softens fabric weave detail. This happens because the color-directive terms in your prompt do not explicitly reinforce the surface properties you want to retain. The fix is to include explicit texture anchors alongside the color directive.

For a knit garment, use: "forest green knit fabric, ribbed texture intact, same stitch pattern, same yarn weight, identical lighting direction from upper left, unchanged background." For a leather bag, use: "burgundy pebbled leather, same grain pattern, same stitching detail, same hardware color, same shadow at base." For a ceramic product, use: "terracotta matte ceramic glaze, same throwing marks on surface, same rim shadow, clean white studio background."

Notice the structure: [new color] + [material] + [specific texture detail] + [hardware or accent colors unchanged] + [lighting preservation] + [background preservation]. This five-part formula prevents the model from simplifying the surface as it recolors. Run a quick test at 512px before committing to full-resolution renders to confirm texture is holding before you launch a full 8-colorway parallel batch.

Once your single recolor node is validated on one colorway, fork the workflow. In the Floniks editor, right-click the source Image Input node and select Duplicate output connection. This creates multiple downstream branches that all receive the same source image. Add one Recolor node per colorway branch, each with its own color-directive prompt. Label each node clearly: "Recolor — Forest Green," "Recolor — Ivory White," "Recolor — Slate Blue."

Connect all branch outputs to a Grid Composer node set to the number of colorways you are producing. The Grid Composer assembles a side-by-side comparison sheet that you can export as a single image for client presentation or design review. Alternatively, connect each branch output to individual Output nodes if you need the variants as separate files for asset management.

For large palettes of 10 or more colorways, consider grouping branches into sets of 5 and running two sequential batches rather than 10 simultaneous branches. Floniks allocates compute across all active branches in a workflow run, and very wide parallel runs on complex image-to-image models can produce longer queue times than two sequential 5-way batches. For palettes of 4 to 8 colorways, full parallelism is optimal and typically completes in under 3 minutes for standard resolution outputs.

AI recoloring is probabilistic, which means a "forest green" prompt does not guarantee a specific hex value. For brand-sensitive work, add a Color Validation node after each recolor branch. This node samples the dominant hue of the output image in the product region and compares it to a target hex value you supply. If the delta-E distance exceeds your tolerance threshold (typically 5–8 for perceptual color difference), the node flags the output for manual review instead of passing it downstream.

When the Color Validation node flags an output, examine the branch prompt. Typical causes of hue drift include the model interpreting your color descriptor as a mixed shade (forest green skewing toward olive), strong background colors bleeding into the product region, or the denoising strength being too high and allowing the model to shift toward its statistical prior for that material type. Corrective actions in order of preference: add a more specific color descriptor ("deep forest green, hex approximately 228B22, not olive, not lime"), reduce denoising strength by 0.05, or switch to the Color Transfer node approach with an explicit swatch.

After validation, route approved outputs to a Brand Archive Output node that appends each colorway file with a structured name: "ProductName_ColorwayName_Resolution_Date." This naming convention feeds cleanly into downstream DAM (digital asset management) systems and makes color variant retrieval instant when a specific colorway needs updating six months later.

The colorway variant workflow is most valuable when it can be applied to any new product image with minimal reconfiguration. Save the parallel-branch structure as a Floniks template, leaving the source Image Input node and each branch color prompt as exposed parameters. When a new product is ready for colorway generation, a team member opens the template, supplies the new hero image, updates the color descriptors for any palette differences, and runs — the full branch architecture and validation logic is already in place.

Document in the template description the three texture anchor categories your team uses most (fabric, leather, ceramic), the validated denoising strength range (0.5–0.65), and the delta-E tolerance for the Color Validation node. This embedded documentation means new team members can operate the workflow without debugging common failure modes from scratch. Version the template when you make structural changes, preserving old versions so that a campaign mid-flight can continue using the validated configuration while a newer version is being tested.