AI Banner Generation: Automating Creatives for Formats

AI-Generated Advertising Banners

A typical situation: a marketer spends half a day manually cutting 15 banners for Yandex.Direct, Google Ads, and VK. Then the designer redoes them because the text didn't fit or the background obscures the product. And so it goes for every new ad cycle. AI-based automation solves this pain: one source yields a hundred ready-made variations in minutes. Our solution focuses on AI banner generation and ad creative automation, supporting multiple banner format generation for Yandex.Direct banners, Google Ads banners, and VK ads banners. We utilize neural networks for advertising, automatic banner cropping, and MLOps for advertising to ensure efficiency. Creative costs drop 5–10 times, and campaign launch speed increases dramatically. We implement a system that generates banners from scratch: background from a prompt (using Flux image generation), text overlay using a grid, adaptation to specific sizes. Everything runs on Python + Flux/DALL-E + custom compositors. With 5+ years of experience and 50+ ad automation projects, we ensure reliable implementation, including banner generation for large e-commerce platforms.

Typical Problems in Banner Creation

Manual cropping for 8+ formats—designers have to manually adapt layouts to every size. We automate it: one prompt yields all sizes simultaneously. Text-background mismatch—the algorithm checks contrast and applies a semi-transparent overlay so the text is readable on any background. Lack of A/B tests—generating dozens of A/B testing banners takes seconds, not days.

How AI Generates Banners for Different Formats

Basic scenario: upload a product image + a text brief (headline, subhead, CTA, brand color). The system generates a background via prompt in a consistent style, then programmatically overlays elements according to proportions. Composer code:

from PIL import Image, ImageDraw, ImageFont
import io

class BannerGenerator:
    STANDARD_SIZES = {
        # Яндекс.Директ
        "yandex_240x400": (240, 400),
        "yandex_300x250": (300, 250),
        "yandex_728x90": (728, 90),
        # Google Ads
        "google_300x250": (300, 250),
        "google_160x600": (160, 600),
        "google_970x250": (970, 250),
        # ВКонтакте
        "vk_1080x607": (1080, 607),
        # Telegram Ads
        "telegram_800x418": (800, 418),
    }

    def __init__(self):
        self.image_gen = FluxImageGenerator()  # или DALL-E / SDXL

    async def generate_banner_set(
        self,
        product_image: bytes,
        headline: str,
        subtext: str,
        cta: str,
        brand_color: str,
        sizes: list[str] = None
    ) -> dict[str, bytes]:
        target_sizes = sizes or list(self.STANDARD_SIZES.keys())
        results = {}

        # Генерируем базовое background изображение
        bg_prompt = f"abstract background, {brand_color} color scheme, modern minimalist, no text, banner design"
        background = await self.image_gen.generate(bg_prompt, width=1920, height=1080)

        for size_name in target_sizes:
            w, h = self.STANDARD_SIZES[size_name]
            banner = self.compose_banner(
                background=background,
                product_image=product_image,
                headline=headline,
                subtext=subtext,
                cta=cta,
                brand_color=brand_color,
                size=(w, h)
            )
            results[size_name] = banner

        return results

    def compose_banner(
        self,
        background: bytes,
        product_image: bytes,
        headline: str,
        subtext: str,
        cta: str,
        brand_color: str,
        size: tuple
    ) -> bytes:
        w, h = size
        bg = Image.open(io.BytesIO(background)).resize(size, Image.LANCZOS)
        canvas = bg.copy()
        draw = ImageDraw.Draw(canvas)

        # Накладываем полупрозрачный оверлей
        overlay = Image.new("RGBA", size, (0, 0, 0, 100))
        canvas = Image.alpha_composite(canvas.convert("RGBA"), overlay).convert("RGB")
        draw = ImageDraw.Draw(canvas)

        # Товар (если горизонтальный баннер — слева, иначе по центру)
        if w > h:  # горизонтальный
            product = Image.open(io.BytesIO(product_image)).convert("RGBA")
            product.thumbnail((h - 20, h - 20))
            canvas.paste(product, (10, (h - product.height) // 2), product.split()[3])
            text_x = h + 10
        else:
            product = Image.open(io.BytesIO(product_image)).convert("RGBA")
            product.thumbnail((w - 20, h // 2))
            canvas.paste(product, ((w - product.width) // 2, 10), product.split()[3])
            text_x = 10

        # Текст
        try:
            font_headline = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", max(12, h // 8))
            font_sub = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", max(10, h // 12))
        except:
            font_headline = ImageFont.load_default()
            font_sub = font_headline

        draw.text((text_x, h // 2), headline, fill="white", font=font_headline)
        draw.text((text_x, h // 2 + h // 8 + 5), subtext, fill="#DDDDDD", font=font_sub)

        # CTA кнопка
        cta_y = h - h // 5
        draw.rounded_rectangle([text_x, cta_y, text_x + w // 3, cta_y + h // 8], radius=5, fill=brand_color)
        draw.text((text_x + 10, cta_y + 5), cta, fill="white", font=font_sub)

        buf = io.BytesIO()
        canvas.save(buf, format="PNG")
        return buf.getvalue()

Why AI Generation Is Faster Than Manual Work

Compare: a designer spends 4–6 hours preparing 10 variations for three formats. An AI system does the same in 2 minutes. Time savings: 120x. Moreover, the number of formats grows to 8+, and A/B tests to 10+ variants. CTR increases by an average of 12% due to automatic headline selection. According to implementation results, creative savings reach 80%, which in monetary terms means a cost of $5 per 100 banners instead of $500. For large campaigns, this translates to thousands of dollars saved monthly.

How to Select the Best Banner Texts

One image is combined with dozens of text variants. Headlines (up to 30 chars), subheads (up to 60), and CTAs (up to 20) are generated via GPT-4o with few-shot examples from your niche. A/B generation code:

async def generate_ab_variants(
    base_brief: dict,
    num_variants: int = 5
) -> list[dict]:
    """Генерируем варианты для A/B тестирования"""
    client = AsyncOpenAI()

    response = await client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "system",
            "content": f"Создай {num_variants} вариантов текста для рекламного баннера. Для каждого: headline (до 30 символов), subtext (до 60 символов), cta (до 20 символов). Верни JSON массив."
        }, {
            "role": "user",
            "content": f"Продукт: {base_brief['product']}\nЦА: {base_brief['audience']}\nОффер: {base_brief['offer']}"
        }],
        response_format={"type": "json_object"}
    )

    variants_text = json.loads(response.choices[0].message.content)["variants"]
    results = []

    for variant in variants_text:
        banners = await generator.generate_banner_set(
            product_image=base_brief["product_image"],
            **variant,
            brand_color=base_brief["brand_color"]
        )
        results.append({"text": variant, "banners": banners})

    return results

Integration with Ad Accounts

class AdPlatformUploader:
    async def upload_to_yandex_direct(self, banners: dict, campaign_id: str): ...
    async def upload_to_vk_ads(self, banners: dict, account_id: str): ...
    async def upload_to_google_ads(self, banners: dict, customer_id: str): ...

Upload occurs via official APIs: Yandex.Direct API and Google Ads API.

Comparison: Manual Cropping vs AI Generation

What's Included in the Ready Solution

Implementation Process

1. Analytics — we study your advertising structure, platform formats, brand book.
2. Design — select a generation model (Flux / DALL-E / SDXL), configure prompts.
3. Implementation — write integration with Pillow, account APIs, A/B module.
4. Testing — run on 100+ real placements, fix artifacts.
5. Deployment — deploy on your infrastructure or cloud, train the team.

Timelines and Cost

• MVP generator with one template and 3 sizes — from 1 to 2 weeks.
• Full system with A/B tests, text variations, and integration — from 3 to 5 weeks.
• Customization for platform specifics (additional formats, brand book) — from 2 weeks.

Cost is calculated individually per task. Assessment takes 1 business day — send a brief. We guarantee correct composition and integration with any platforms. Order a consultation on implementing AI banner generation. Contact us for a system demonstration.

Generative AI Development: From Prompt to Production API

We often receive a task "generate a product image" — on the surface it seems simple. But behind this lies a choice between dozens of models, configuring the inference pipeline, manually solving consistency issues, integrating into the product backend, and answering why the model generates hands with six fingers in staging but not in production. Let's break down the directions we work with.

Image Generation: From Prompt to Production API

The current landscape includes FLUX.1 [dev/schnell/pro] from Black Forest Labs and Stable Diffusion 3.5. FLUX.1 [schnell] takes 4 steps instead of 20–50 for SDXL — 5–12 times faster — while maintaining higher quality. On an A100 80GB — 1.2–1.8 s per 1024×1024 image at batch_size=4.

A typical deployment issue: FLUX.1 [dev] requires 24+ GB VRAM in fp16. On A10G 24GB it fits tightly; at batch_size>1 — OOM. Solution: torch_dtype=torch.bfloat16 + enable_model_cpu_offload() from diffusers, or quantization via bitsandbytes to NF4 — minimal quality drop, memory consumption drops to 12–14 GB.

ControlNet and IP-Adapter are key tools for production tasks where controllability is needed. ControlNet with Canny/Depth/Pose maps provides structural control. IP-Adapter (especially IP-Adapter-FaceID) allows transferring character identity to generations — this is the foundation for personalized content. More about ControlNet can be found on Wikipedia.

Case study: e-commerce photography. A retailer with 8000 SKUs needed lifestyle photos for each product. Pipeline: product segmentation (Segment Anything Model 2) → background removal → inpainting with FLUX.1 [dev] using product image as IP-Adapter reference → upscale via RealESRGAN_x4plus. The generation cost is negligible compared to professional photography, providing huge savings. Throughput — 200 images/hour on 2× A100. Our extensive experience from 30+ projects ensures we select the optimal model for your task — an evaluation can be obtained upfront.

Why Is Model Selection Only Half the Battle?

Fine-tuning for a Specific Style or Character

Dreambooth and LoRA are the standard for adapting to a specific visual style or object. LoRA trains in 2–4 hours on 20–30 reference images on a single A100. Rank 16–32 is usually sufficient for style; rank 64+ is needed for precise face reproduction.

A common mistake: training LoRA too long — the model overfits to references, losing the ability to vary. Sign: at cfg_scale=7, all images look like copy-paste of references. Solved by early stopping (usually 1500–2000 steps for 20 images) and prior_preservation_loss.

For deeper customization — full fine-tuning via diffusers + accelerate with FSDP on multiple GPUs. But that already takes 40–80 hours of training and requires a truly large dataset (1000+ images).

Comparison of Image Generation Approaches

Video Generation: Which Models Are Best?

Open-weight video models still lag behind commercial ones in stability and length. Wan2.1 is the best choice for self-hosting: 14B parameters, runs on 2× A100, delivers acceptable quality for short clips.

The main pain of video generation is temporal consistency: the character changes clothing color at the third second, objects "drift." Partial solution — generation with motion_bucket_id and noise_aug_strength in Stable Video Diffusion, or using I2V (image-to-video) instead of pure text-to-video. As noted in VideoPoet research, consistency is achieved by training on long sequences.

AnimateDiff remains a working tool for short loops and motion effects on top of SD/FLUX. Not Sora, but deployable locally and predictable.

Music and Audio Generation

AudioCraft from Meta (MusicGen + AudioGen) is a production-ready stack for music generation. musicgen-large (3.3B) generates 30 s of music in ~8 s on A100. Control via text prompt and melody conditioning — you can specify a melody by humming.

Stable Audio Open from Stability AI is an alternative with length up to 47 s, better structural control (intro/verse/chorus). Deployment is similar: diffusers + FastAPI.

For voice-over and dubbing — ElevenLabs API or self-hosted XTTS v2 (see Speech AI service). For sound design and foley — AudioGen.

3D Generation: Current Practical State

3D generation has not yet reached the same maturity as 2D. But for specific tasks, tools are already working:

TripoSG and Shap-E — text/image-to-3D. Shap-E from OpenAI generates simple 3D meshes in seconds, but geometry is rough. TripoSG gives more detailed results but requires post-processing (remeshing, UV unwrapping).

Wonder3D and Zero123++ — 3D reconstruction from a single image. They work by generating multi-views (6–8 views) and then 3D reconstruction via NeuS or instant-ngp.

Gaussian Splatting (3DGS) — not generation, but reconstruction from a series of photos/videos. For product cards and real estate it's already production: 50–200 photos → 3DGS model in 15–30 min on RTX 4090 → interactive 3D viewer in browser.

What Infrastructure Is Needed for Generative AI Deployment?

Critical for generative models:

• Task queue — Celery + Redis or Ray Serve. Synchronous HTTP for image generation is unacceptable with >5 concurrent requests.
• Caching — similar prompts yield similar results. Semantic cache via embeddings (faiss + sentence-transformers) can reduce GPU load by 20–40%.
• Quality monitoring — CLIP score for text-image alignment, FID for evaluating generation distribution. Integrate into MLflow or Weights & Biases.
• Storage — generated images immediately to S3/MinIO, not on the inference server disk.

What's Included in the Deliverables

We take the project turnkey — from model selection to deployment and monitoring. The result includes:

• Model (or API integration) with performance benchmarks (latency p99, throughput).
• Pipeline documentation (prompt engineering guide, model card, dependency versions).
• Integration with your backend (REST/gRPC, queues).
• Configured monitoring (dashboards, alerts for quality drift).
• Training workshop for the team (2–4 hours).
• Warranty support for 3 months after launch — as part of our quality certificate.

We have completed 30+ projects in generative AI — this gives us the right to guarantee results.

How Is the Generative AI Development Process Structured?

1. Analysis (1–2 days): audit of current architecture, clarification of use case, selection of models and success metrics. We evaluate the project free of charge.
2. Proof of Concept (1–3 weeks): quick prototype on your data — to see real quality, not blog demos.
3. Design (1–2 weeks): pipeline architecture, infrastructure (GPU cluster/API), A/B testing plan.
4. Implementation and fine-tuning (4–12 weeks): development, LoRA/full fine-tuning, integration with queue and cache.
5. Testing (1–2 weeks): load tests, metric validation, edge-case verification (negative scenarios).
6. Deployment and monitoring (1–2 weeks): production deployment, monitoring setup, documentation.

Timeline Estimates

Integration of a ready API (DALL·E 3, Midjourney API, Stability API) — 1–2 weeks. Self-hosted pipeline with fine-tuning — 6–12 weeks. Full platform with UI, queues and monitoring — 3–6 months. The specific cost is calculated individually after analyzing your scenario.

Contact us — order a consultation, and we will select the optimal architecture for your project. Get a preliminary cost and timeline estimate for free.