A Multi-Format Export Workflow (One Render, Every Aspect Ratio)

The naive approach to delivering assets in multiple formats is to run the same prompt several times and change the aspect ratio parameter each time. This seems straightforward but introduces three compounding problems. First, it multiplies credit cost linearly with the number of target formats — four platforms means four full generations. Second, each generation is statistically independent, so the subject's pose, expression, and the precise distribution of background elements will differ between runs even with the same seed. Third, diffusion-based models do not guarantee that a 16:9 composition will crop cleanly to a 9:16 output — the subject may be partially cut off in the vertical variant, requiring manual intervention anyway.

A multi-format export workflow solves all three problems by separating the generation step from the formatting step. You generate once at a generous square or panoramic resolution that contains enough visual information for all target crops. Then you apply deterministic, automated crop-and-resize logic in parallel downstream nodes. The result is perfectly consistent color grading, identical subjects, and zero per-format re-prompting. This architecture is also faster in wall-clock time because the crop nodes run in parallel after the single upstream generation completes.

The master generation node should produce an image at a resolution and aspect ratio that fully contains the composition required by every downstream format. For a workflow targeting 16:9 widescreen, 9:16 vertical, and 1:1 square simultaneously, a 2:1 panoramic master (for example 2048×1024) gives the horizontal breadth needed for 16:9 while still containing a centered square crop for 1:1. If your narrowest target is 4:5, ensure the master is tall enough to preserve headroom and footroom after cropping.

Write your generation prompt with center-weighted composition in mind: "subject centered, equal negative space left and right, safe zone 300px from each edge." This prompt instruction tells the model to keep the primary subject away from the frame boundaries, ensuring every crop variant retains it. Add negative prompt tokens such as "subject at edge, clipped face, cut-off limbs" to further reinforce center composition. Set the model's guidance scale moderately high (7–9 for most diffusion models) so the composition instruction is followed faithfully rather than being overridden by stylistic variation.

After the master generation node, connect its output port to a series of parallel Crop and Resize nodes — one per target format. In the Floniks editor at /editor, you can hold Shift and click multiple destination ports from a single source node to draw batch connections in one gesture. Each crop node should be configured with a target aspect ratio, a crop anchor point (center-top for portrait compositions where faces must be retained at the top of the frame, center-center for product and lifestyle shots), and an output resolution matching the platform specification.

Typical platform configurations: YouTube thumbnail at 1280×720 (16:9, crop anchor center-center), Instagram Reels at 1080×1920 (9:16, crop anchor center-top to preserve faces), Instagram feed at 1080×1080 (1:1, center-center), Facebook Ads at 1080×1350 (4:5, center-center). Label each crop node with the platform name so the canvas is self-documenting. Finally, route each crop node's output to a named Output Collector node labeled with the platform so downstream users can immediately identify which file to download without opening metadata.

A fixed crop anchor works well for most compositions, but fails when the subject shifts significantly from center — common in dynamic lifestyle shots or multi-subject compositions. To handle this, insert a Focal Point Detection node between the master generation node and the fanout crop layer. This node analyzes the generated image using a saliency map or face-detection pass and returns a focal point coordinate (x, y as fractions of image width and height). Each downstream crop node reads this coordinate as its dynamic anchor instead of a fixed center offset.

When the focal point detector identifies a face at (0.65, 0.3) — upper-right quadrant — every crop node adjusts its anchor accordingly, ensuring the face stays in-frame across all formats. This is particularly important for 9:16 vertical crops, where a face detected in the center-left of a 16:9 master would be cropped out by a naive center anchor. The focal point detector adds one lightweight processing step but eliminates the category of "wrong crop" failures that would otherwise require manual correction per format.

Once the workflow is validated end-to-end, save it as a named template. The template captures the master generation node's model and prompt structure, the focal point detector configuration, and all four (or more) crop node definitions. Future campaigns start by opening the template, updating the prompt in the master generation node, and running. No downstream configuration changes are needed unless a new platform format is added.

Credit cost for this workflow is equivalent to a single generation at the master resolution — the crop and resize nodes are deterministic image-processing operations that do not invoke external AI models and therefore do not consume generation credits. This means delivering assets in five formats costs exactly the same as delivering one, making the multi-format export workflow one of the highest leverage patterns in the Floniks template library for teams with cross-platform publishing needs.